Intelligent Network Design Driven by Big Data

Analytics, IoT, AI and Cloud Computing

Intelligent Network Design

As enterprise access networks evolve with a larger number of mobile users, a wide range of

Analytics, IoT, AI and Cloud Computing

Intelligent Network Design Driven by Big Data
devices and new cloud-based applications, managing user performance on an end-to-end
basis has become rather challenging. Recent advances in big data network analytics combined
with AI and cloud computing are being leveraged to tackle this growing problem. AI is
becoming further integrated with software that manage networks, storage, and can compute. Driven by Big Data Analytics,
IoT, AI and Cloud Computing
This edited book focuses on how new network analytics, IoTs and Cloud Computing platforms
are being used to ingest, analyse and correlate a myriad of big data across the entire network
stack in order to increase quality of service and quality of experience (QoS/QoE) and to
improve network performance. From big data and AI analytical techniques for handling the
huge amount of data generated by IoT devices, the authors cover cloud storage optimization,
the design of next generation access protocols and internet architecture, fault tolerance and
reliability in intelligent networks, and discuss a range of emerging applications.

This book will be useful to researchers, scientists, engineers, professionals, advanced students Edited by
and faculty members in ICTs, data science, networking, AI, machine learning and sensing. It
will also be of interest to professionals in data science, AI, cloud and IoT start-up companies, Sunil Kumar, Glenford Mapp and Korhan Cengiz
as well as developers and designers.

About the Editors

Sunil Kumar is an associate professor of computer science and engineering at Amity
University, Noida campus, India.

Glenford Mapp is an associate professor at Middlesex University, London, UK.

Korhan Cengiz is an assistant professor of electrical and electronics engineering at Trakya

University, Turkey.

Edited by Kumar, Mapp and Cengiz

The Institution of Engineering and Technology

theiet.org
978-1-83953-533-8
Intelligent Network Design
Driven by Big Data Analytics,
IoT, AI and Cloud Computing
Other volumes in this series:

Volume 1 Knowledge Discovery and Data Mining M.A. Bramer (Editor)

Volume 3 Troubled IT Projects: Prevention and turnaround J.M. Smith
Volume 4 UML for Systems Engineering: Watching the wheels, 2nd Edition J. Holt
Volume 5 Intelligent Distributed Video Surveillance Systems S.A. Velastin and P. Remagnino (Editors)
Volume 6 Trusted Computing C. Mitchell (Editor)
Volume 7 SysML for Systems Engineering J. Holt and S. Perry
Volume 8 Modelling Enterprise Architectures J. Holt and S. Perry
Volume 9 Model-Based Requirements Engineering J. Holt, S. Perry and M. Bownsword
Volume 13 Trusted Platform Modules: Why, when and how to use them A. Segall
Volume 14 Foundations for Model-based Systems Engineering: From Patterns to Models J. Holt, S. Perry and
M. Bownsword
Volume 15 Big Data and Software Defined Networks J. Taheri (Editor)
Volume 18 Modeling and Simulation of Complex Communication M. A. Niazi (Editor)
Volume 20 SysML for Systems Engineering: A Model-Based Approach, 3rd Edition J. Holt and S. Perry
Volume 22 Virtual Reality and Light Field Immersive Video Technologies for Real-World Applications G. Lafruit
and M. Tehrani
Volume 23 Data as Infrastructure for Smart Cities L. Suzuki and A. Finkelstein
Volume 24 Ultrascale Computing Systems J. Carretero, E. Jeannot and A. Zomaya
Volume 25 Big Data-Enabled Internet of Things M. Khan, S. Khan, A. Zomaya (Editors)
Volume 26 Handbook of Mathematical Models for Languages and Computation A. Meduna, P. Horáček, M. Tomko
Volume 29 Blockchains for Network Security: Principles, technologies and applications H. Huang, L. Wang, Y. Wu,
K. R. Choo (Editors)
Volume 32 Network Classification for Traffic Management: Anomaly detection, feature selection, clustering and
classification Z. Tari, A. Fahad, A. Almalawi and X. Yi
Volume 33 Edge Computing: Models, technologies and applications J.Taheri and S. Deng (Editors)
Volume 34 AI for Emerging Verticals: Human-robot computing, sensing and networking M. Z. Shakir and N. Ramzan
(Editors)
Volume 35 Big Data Recommender Systems Vol 1 & 2 O. Khalid, S. U. Khan, A. Y. Zomaya (Editors)
Volume 37 Handbook of Big Data Analytics Vol 1 & 2 V. Ravi and A. K. Cherukuri (Editors)
Volume 39 ReRAM-based Machine Learning H. Y, L. Ni and S. M. P. Dinakarrao
Volume 40 E-learning Methodologies: Fundamentals, technologies and applications M. Goyal, R. Krishnamurthi and
D. Yadav (Editors)
Volume 44 Streaming Analytics: Concepts, architectures, platforms, use cases and applications P. Raj,
C. Surianarayanan, K. Seerangan and G. Ghinea (Editors)
Volume 44 Streaming Analytics: Concepts, architectures, platforms, use cases and applications P. Raj, A. Kumar,
V. García Díaz and N. Muthuraman (Editors)
Volume 115 Ground Penetrating Radar: Improving sensing and imaging through numerical modelling
X. L. Travassos, M. F. Pantoja and N. Ida
Intelligent Network Design
Driven by Big Data Analytics,
IoT, AI and Cloud Computing
Edited by
Sunil Kumar, Glenford Mapp and Korhan Cengiz

The Institution of Engineering and Technology

Published by The Institution of Engineering and Technology, London, United Kingdom
The Institution of Engineering and Technology is registered as a Charity in England &
Wales (no. 211014) and Scotland (no. SC038698).
© The Institution of Engineering and Technology 2022
First published 2022

This publication is copyright under the Berne Convention and the Universal Copyright
Convention. All rights reserved. Apart from any fair dealing for the purposes of research or
private study, or criticism or review, as permitted under the Copyright, Designs and Patents
Act 1988, this publication may be reproduced, stored or transmitted, in any form or by
any means, only with the prior permission in writing of the publishers, or in the case of
reprographic reproduction in accordance with the terms of licences issued by the Copyright
Licensing Agency. Enquiries concerning reproduction outside those terms should be sent to
the publisher at the undermentioned address:
The Institution of Engineering and Technology
Futures Place
Kings Way, Stevenage
Herts, SG1 2UA, United Kingdom
www.theiet.org
While the authors and publisher believe that the information and guidance given in this
work are correct, all parties must rely upon their own skill and judgement when making use
of them. Neither the author nor publisher assumes any liability to anyone for any loss or
damage caused by any error or omission in the work, whether such an error or omission is
the result of negligence or any other cause. Any and all such liability is disclaimed.
The moral rights of the author to be identified as author of this work have been asserted by
him in accordance with the Copyright, Designs and Patents Act 1988.

British Library Cataloguing in Publication Data

A catalogue record for this product is available from the British Library

ISBN 978-1-83953-533-8 (hardback)

ISBN 978-1-83953-534-5 (PDF)

Typeset in India by Exeter Premedia Services Private Limited

Printed in the UK by CPI Group (UK) Ltd, Croydon
Cover Image: Yuichiro Chino via Getty Images
 Contents

About the Editors xv

1 
Introduction to intelligent network design driven by big data analytics,
IoT, AI and cloud computing 1
Sunil Kumar , Glenford Mapp , and Korhan Cengiz 
Preface1
Chapter 2:   Role of automation, Big Data, AI, ML IBN, and
cloud computing in intelligent networks 2
Chapter 3:   An intelligent verification management approach for
efficient VLSI computing system 2
Chapter 4:   Evaluation of machine learning algorithms on academic
big dataset by using feature selection techniques 3
Chapter 5:   Accurate management and progression of Big Data analysis 4
Chapter 6:   Cram on data recovery and backup cloud computing techniques 4
Chapter 7:   An adaptive software defined networking (SDN) for load
balancing in cloud computing 5
Chapter 8:   Emerging security challenges in cloud computing: An insight 5
Chapter 9:   Factors responsible and phases of speaker recognition system 6
Chapter 10: IoT-based water quality assessment using fuzzy logic controller 6
Chapter 11: Design and analysis of wireless sensor network for
intelligent transportation and industry automation 7
Chapter 12: A review of edge computing in healthcare Internet of
Things: theories, practices, and challenges  7
Chapter 13: Image processing for medical images on the basis of
intelligence and bio computing 8
Chapter 14: IoT-based architecture for smart health-care systems 8
Chapter 15: IoT-based heart disease prediction system 9
Chapter 16: DIAIF: detection of interest flooding using artificial
intelligence-based framework in NDN android 9
Chapter 17: Intelligent and cost-effective mechanism for
monitoring road quality using machine learning 10
References10
vi  Intelligent network design driven by Big Data analytics

2 
Role of automation, Big Data, AI, ML IBN, and cloud computing in
intelligent networks13
Sunil Kumar  and Priya Ranjan
2.1 Evolution of networks: everything is connected  13
2.1.1 Intelligent devices  14
2.1.2 Intelligent devices connection with networks  14
2.2 Huge volume of data generation by intelligent devices  15
2.2.1 Issues and challenges of Big Data Analytics  16
2.2.2 Storage of Big Data  16
2.3 Need of data analysis by business  18
2.3.1 Sources of information  19
2.3.2 Data visualization  20
2.3.3 Analyzing Big Data for effective use of business  20
2.3.4 Intelligent devices thinking intelligently  20
2.4 Artificial intelligence and machine learning in networking  21
2.4.1 Role of ML in networks  21
2.5 Intent-based networking  22
2.6 Role of programming  23
2.6.1 Basic programming using Blockly  23
2.6.2 Blockly games  24
2.7 Role of technology to design a model  24
2.7.1 Electronic toolkits  25
2.7.2 Programming resources  26
2.8 Relation of AI, ML, and IBN  26
2.9 Business challenges and opportunities  26
2.9.1 The evolving job market  27
2.10 Security  28
2.10.1 Challenges to secure device and networks  28
2.11 Summary  31
References  31

3 
An intelligent verification management approach for efficient VLSI
computing system  35
Konasagar Achyut , Swati K Kulkarni , Akshata A Raut , Siba Kumar Panda ,
and Lakshmi Nair  
3.1 Introduction  36
3.2 Literature study  38
3.3 Verification management approach: Case Study 1  46
3.3.1 The pseudo random number generator in a verification
environment  47
3.3.2 Implementation of PRNG in higher abstraction language and
usage of DPI  49
3.4 Verification management approach: Case Study 2  51
 Contents  vii

3.5 Challenges and research direction  54

3.5.1 Challenges in intelligent systems  54
3.6 Conclusion  55
References  55

4 
Evaluation of machine learning algorithms on academic big dataset by
using feature selection techniques  61
Mukesh Kumar , Amar Jeet Singh , Bhisham Sharma , and Korhan Cengiz
4.1 Introduction  62
4.1.1 EDM  64
4.1.2 EDM process  65
4.1.3 Methods and techniques  66
4.1.4 Application areas of data mining  68
4.2 Literature survey  69
4.3 Materials and methods  72
4.3.1 Dataset description  72
4.3.2 Classification algorithms  73
4.3.3 FS algorithms  75
4.3.4 Data preprocessing phase  77
4.4 Implementation of the proposed algorithms  78
4.4.1 Model construction for the standard classifier  78
4.4.2 Implementation after attribute selection using ranker method  79
4.5 Result analysis and discussion  84
4.6 Conclusion  86
References  86

5 Accurate management and progression of Big Data Analysis  93

Tanmay Sinha, Narsepalli Pradyumna , and K B Sowmya 
5.1 Introduction  93
5.1.1 Examples of Big Data  94
5.2 Big Data Analysis  96
5.2.1 Life cycle of Big Data  96
5.2.2 Classification of the Big Data  97
5.2.3 Working of Big Data Analysis  99
5.2.4 Common flaws that undermine Big Data Analysis  100
5.2.5 Advantages of Big Data Analysis  100
5.3. Processing techniques  101
5.3.1 Traditional method  101
5.3.2 MapReduce  102
5.3.3 Advantages of MapReduce  103
5.4 Cyber crime  105
5.4.1 Different strategies in Big Data to help in various
circumstances  106
viii  Intelligent network design driven by Big Data analytics

5.4.2 Big Data Analytics and cybercrime  107

5.4.3 Security issues associated with Big Data  108
5.4.4 Big Data Analytics in digital forensics  108
5.5 Real-time edge analytics for Big Data in IoT  109
5.6 Conclusion  111
References  112

6 Cram on data recovery and backup cloud computing techniques  115

 Dharanyadevi P , Julie Therese M , Senthilnayaki B ,
Devi A , and Venkatalakshmi K
6.1 Introduction  115
6.1.1 Origin of cloud  116
6.1.2 Sole features of cloud computing  116
6.1.3 Advantages of cloud computing  116
6.1.4 Disadvantages of cloud computing  117
6.2 Classification of data recovery and backup  118
6.2.1 Recovery  118
6.2.2 Backup  119
6.3 Study on data recovery and backup cloud computing techniques  120
6.3.1 Backup of real-time data and recovery using cloud computing 120
6.3.2 Data recovery and security in cloud  122
6.3.3 Amoeba: An autonomous backup and recovery solid-state
drives for ransomware attack defense  123
6.3.4 A cloud-based automatic recovery and backup system for
video compression  124
6.3.5 Efficient and reliable data recovery techniques in
cloud computing  126
6.3.6 Cost-efficient remote backup services for enterprise cloud  127
6.3.7 DR-cloud: Multi-cloud-based disaster recovery service  128
6.4 Conclusion  131
References  131

7 
An adaptive software-­defined networking (SDN) for load balancing in
cloud computing  135
Swati Lipsa , Ranjan Kumar Dash , and Korhan Cengiz 
7.1 Introduction  135
7.2 Related works  138
7.3 Architecture overview of SDN  140
7.4 Load-balancing framework in SDN  141
7.4.1 Classification of SDN controller architectures  142
7.5 Problem statement  144
7.5.1 Selection strategy of controller head  144
7.5.2 Network setup  146
7.6 Illustration  147
 Contents  ix

7.7 Results and discussion  150

7.7.1 Comparison of throughput  150
7.7.2 Comparison of PTR  151
7.7.3 Comparison of number of packet loss  152
7.8 Conclusion  152
References  153

8 Emerging security challenges in cloud computing: an insight  159

 Gaurav Aggarwal , Kavita Jhajharia , Dinesh Kumar Saini , and Mehak Khurana 
8.1 Introduction  159
8.1.1 An introduction to cloud computing and its security  159
8.2 The security issues in different cloud models and threat management
techniques  161
8.2.1 Five most indispensable characteristics  161
8.2.2 The security issues in cloud service model  162
8.2.3 Security issues in cloud deployment models  165
8.2.4 Security challenges in the cloud environment  166
8.2.5 The threat management techniques  169
8.3 Review on existing proposed models  174
8.3.1 SeDaSC  174
8.3.2 The ‘SecCloud’ protocol  176
8.3.3 Data accountability and auditing for secure cloud data storage 179
8.4 Conclusion and future prospectives  180
References  180

9 Factors responsible and phases of speaker recognition system  185

Hunny  and Ayush Goyal 
9.1 Study of related research  185
9.2 Phases of speaker recognition system  187
9.2.1 Speaker database collection  187
9.2.2 Feature extraction  192
9.2.3 Feature mapping  195
9.3 Basics of speech signals  197
9.3.1 Speech production system  197
9.3.2 Speech perception  198
9.3.3 Speech signals  198
9.3.4 Properties of the sinusoids  198
9.3.5 Windowing signals  199
9.3.6 Zero-crossing rate  199
9.3.7 Autocorrelation  200
9.4 Features of speech signals  200
9.4.1 Physical features  200
9.4.2 Perceptual features  202
9.4.3 Signal features  202
x  Intelligent network design driven by Big Data analytics

9.5 Localization of speaker  203

9.6 Conclusion  203
References  203

10 IoT-­based water quality assessment using fuzzy logic controller  213

 A. Hendry , K. Ghousiya Begum , A. Hariharan , and K. J. Aravind 
10.1 Introduction  213
10.2 Experimental procedures  217
10.3 Working  220
10.4 Results and discussions  227
10.5 Conclusion  228
References  231

11 Design and analysis of wireless sensor network for intelligent

transportation and industry automation 235
Prabhakar D. Dorge , Prasanna M. Palsodkar , and Divya Dandekar 
11.1 Introduction 235
11.2 Wireless sensor network 245
11.3 WSN application 246
11.4 Limitations of WSN 247
11.5 Literature survey 247
11.6 Related work 247
11.7 Methodology 248
11.7.1 Throughput 248
11.7.2 Delay 248
11.7.3 Packet delivery ratio 249
11.7.4 Design of WiMAX-based WSN system 249
11.8 Related results 250
11.9 Conclusion 251
11.10 Future scope 252
References253

12 A review of edge computing in healthcare Internet of things: theories,

practices and challenges 257
Shamik Tiwari  and Vadim Bolshev 
12.1 Introduction 258
12.2 Cloud computing in healthcare and its limitations 258
12.2.1 Public cloud 259
12.2.2 Private cloud 260
12.2.3 Hybrid cloud 260
12.2.4 Community cloud 260
12.3 Edge computing and its advantages over cloud computing 261
 Contents  xi

12.3.1 Advantages of edge/fog computing 261

12.3.2 Disadvantages of edge/fog computing 262
12.4 IoT in healthcare 263
12.5 Edge computing in healthcare 264
12.6 Machine learning in healthcare 265
12.7 Integrated role of IOT, ML and edge computing in healthcare 268
12.7.1 Patient care during surgical procedure 269
12.7.2 Patient care at home 269
12.7.3 Patient care in ambulance 269
12.8 Modelling and simulation tools for edge/fog computing 269
12.9 Edge computing in Covid-19 pandemic era 271
12.10 Challenges of edge computing 272
12.11 Conclusion 273
References274

13 Image Processing for medical images on the basis of intelligence and 

biocomputing279
M. Mohammed Mustafa , S. Umamaheswari , and Korhen Cengiz
13.1 Introduction 279
13.1.1 What is an image? 280
13.2 Image processing 280
13.2.1 Equivalent image processing 280
13.2.2 Digital image processing 281
13.2.3 Digital image 281
13.2.4 Applications of color models 281
13.2.5 Applications of digital image processing 281
13.2.6 Fundamental steps in digital image processing 282
13.2.7 Components of an image processing system 283
13.3 Medical imaging 284
13.4 Deep learning techniques 284
13.4.1 Uses of image processing 287
13.5 Convolutional neural network 288
13.6 Convolution layers 289
13.6.1 Training phase 292
13.6.2 Training strategies 292
13.6.3 CNN performance 293
13.6.4 Convolutional neutral networks with AI 296
13.6.5 CNN layers 296
13.6.6 CNN image classifier 297
13.7 Deep learning for lung disease detection 298
13.7.1 Preprocessing of images 298
13.7.2 Training 299
13.7.3 Classification 299
13.8 Conclusion 300
References300
xii  Intelligent network design driven by Big Data analytics

14 IoT-­based architecture for smart health-­care systems 303

 S P Rajamohana , Zbigniew M Wawrzyniak , T A Deva Priyan , V Rajeevan ,
N Vaibhav Ram , J Vinith , J Keerthanaa , and Priyanka Raj

14.1 Introduction 303
14.2 Literature survey 305
14.3 Related works 309
14.4 Hardware components and sensors 312
14.4.1 Development boards 312
14.4.2 Sensors 312
14.4.3 Other modules 314
14.5 Proposed work 315
14.5.1 Hardware components used 316
14.6 Implementation and results 320
14.7 Conclusion 323
References323

15 IoT-­based heart disease prediction system 329

 Rajamohana S P, Zbigniew M Wawrzyniak , Krishna Prasath S, Shevannth R ,
Raja Kumar I, Mohammed Rafi M , and T Hariprasath D

15.1 Introduction 329
15.1.1 Deep learning 330
15.2 Related work 332
15.3 Proposed system 344
15.3.1 Arduino UNO 345
15.3.2 Heartbeat sensor 346
15.3.3 Temperature sensor 347
15.3.4 Pressure sensor 347
15.3.5 Liquid crystal display (LCD) display (16 × 2) 347
15.4 Advantages of proposed system 350
15.5 Limitations of proposed system 350
15.6 Results and discussion 350
15.7 Conclusion 351
References352

16 DIAIF: Detection of Interest Flooding using Artificial Intelligence-­based

Framework in NDN android 357
P Vimala Rani, Rajkumar , Narshimha Malika Arjunan , and S Mercy Shalinie

16.1 Introduction 357
16.2 Background 358
16.2.1 ICN communication model 359
16.2.2 Generating IFA in NDN android 359
16.3 Proposed methodology: RD Iterative Adaptive Inverse Filtering
(RD-IAIF)360
16.3.1 Attack detection 361
 Contents  xiii

16.4 Real-time deployment on emergency applications 362

16.4.1 Communication establishment through NFD 363
16.4.2 NDN application on android 365
16.4.3 Attack detection on NFD android using AI 368
16.5 Conclusion 369
References369

17 Intelligent and cost-effective mechanism for monitoring road

quality using machine learning 371
Prabhat Singh, Abhay Bansal, Ahmad E Kamal, and Sunil Kumar
17.1 Introduction 372
17.1.1 Definition of TMS 372
17.1.2 Issues and challenges faced in TMS 372
17.2 Literature review of machine learning for road condition detection 378
17.3 Gaps identified in the literature 385
17.4 Proposed methodology 385
17.5 Implementation 387
17.5.1 Flutter-based application 387
17.5.2 Machine learning models 388
17.6 Results 391
17.6.1 Comparison with the existing models 391
17.7 Conclusion and future scope 394
References 394

18 Conclusion397
Sunil Kumar , Glenford Mapp , and Korhan Cengiz
18.1 Conclusion 397
References397

Index  401
This page intentionally left blank
 About the Editors

Sunil Kumar is an associate professor of Computer Science and Engineering at

Amity University, Noida campus, India. His research interests include computer
networks, distributed systems, wireless sensor networks, SDN, and big data. He
is industry CCNA & CCNP certified. He is a member of the IET, CSTA, IAER,
IAENG. He holds a PhD in energy optimization in distributed wireless sensor net-
works from Amity University, Noida India.

Glenford Mapp is an associate professor at Middlesex University, London, UK.

His primary expertise is in the development of new technologies for mobile and
distributed systems such as service platforms, cloud computing, network addressing
and transport protocols for local environments. He had previously worked for AT&T
Cambridge Laboratories for ten years. He received his PhD in computer science
from the University of Cambridge, UK.

Korhan Cengiz is an assistant professor of electrical and electronics engineering

at Trakya University, Turkey. His research interests include computer networks, big
data, wireless sensor networks, wireless communications, routing protocols, statisti-
cal signal processing, indoor positioning systems, power electronics and machine
learning. He is an associate editor of Interdisciplinary Sciences: Computational Life
Sciences, handling editor of Microprocessors and Microsystems, and associate edi-
tor of IET Electronics Letters, IET Networks, amongst others.
This page intentionally left blank
 Chapter 1
Introduction to intelligent network design
driven by big data analytics, IoT, AI and
cloud computing
Sunil Kumar 1, Glenford Mapp 2, and Korhan Cengiz 3

Preface

As enterprise access networks evolve with more mobile users, diverse devices and
cloud-­based applications, managing user performance on an end-­to-­end basis has
become next to impossible. Recent advances in big data network analytics, com-
bined with AI and cloud computing are being leveraged to tackle this growing prob-
lem. The book focuses on how new network analytics platforms are being used to
ingest, analyze and correlate a myriad of infrastructure data across the entire net-
work stack with the goal of finding and fixing the quality of service network perfor-
mance problems.
This book presents new upcoming technologies in the field of networking and
telecommunication. It addresses major new technological developments and reflects
on industry needs, current research trends and future directions. The authors focus
on the development of AI-­powered mechanisms for future wireless networking
applications and architectures which will lead to more performant, resilient and
valuable ecosystems and automated services. The book is a primary readership and
is a “must-­read” for researchers, academicians, engineers and scientists involved in
the design and development of protocols and AI applications for wireless communi-
cation devices and wireless networking technologies.
All chapters presented here are the product of extensive field research involving
applications and techniques related to data analysis in general, and to big data, AI,
IoT and network technologies in particular.

1
Department of Computer Science & Engineering, Amity University, India
2
Faculty of Science & Technology, Department of Computer Science, Middlesex University, London,
UK
3
Department of Electrical-­Electronics Engineering, Trakya University, Edirne, Turkey
2  Intelligent network design driven by Big Data analytics

Chapter 2: Role of automation, Big Data, AI, ML IBN, and cloud

computing in intelligent networks
There are more smart devices in our world today than individuals. A growing num-
ber of people are linked to the Internet 24 hours a day, in some form or another. A
growing number of people own three, four, or more smart devices and rely on them.
Smartphones, fitness and health trackers, e-­readers, and tablets are just a few exam-
ples. It is forecast that on average there will be 3.4 smart devices or connections for
every person on earth [1, 2]. The Internet of Things (IoT) is relevant to many indus-
tries. IoT systems contribute to the environmental controls, retail, transportation,
healthcare, and agriculture industries among many others. According to Statista, the
number of IoT devices that are in use across all relevant industries is forecast to
grow to more than 8 billion by 2030 [3, 4]. As for consumers, important growth
areas are the Internet and digital media devices, which include smartphones. This
area is also predicted to grow to more than 8 billion by 2030 [5]. Other applications
with more than 1 million connected devices are connected and autonomous vehicles,
IT infrastructure, asset management, and electric utility smart grid [6].
All of this is made possible through intelligent networks. The planet is rap-
idly becoming covered in networks that allow digital devices to communicate and
interconnect. Consider the network mesh as a digital skin that surrounds the earth.
Mobile devices, electronic sensors, electronic measuring equipment, medical gad-
gets, and gauges can all link with this digital skin. They keep track of, communicate
with, analyze, and, in some situations, automatically adjust to the data collected and
transmitted [7, 8].
Keywords: AI, ML, cloud computing, IBN, Big Data, IoT, automation, virtual-
ization, intelligent networks.

Chapter 3: An intelligent verification management approach for

efficient VLSI computing system
Any masterpiece is conjoined with all works of engineering, which includes the field
of computer science or an electrical and electronics or mixture of both computer and
electronics. Today, this gives the industry to understand research, evolve and develop
into newer technology unfolding many scriptures behind the engineering works.
In the similar manner, this chapter unfolds the prominent works involved in the
verification of the designs involved in VLSI domain. Considering machine learning
(ML), neural networks and artificial intelligence (AI) concepts and applying these to
a wide range of verification approaches are quite interesting. The specific kinds of
Register Transfer Level (RTL) design require rigorous verification which is targeted
over any type of Field Programmable Gate Array (FPGA) or application-­specific
integrated circuits (ASICs). The verification process should be closed with testing
all possible scenarios that too with intelligent verification methods. This chapter
in the following pages brings the unique way of verification procedure involved in
the RTL development methodologies using hardware description languages. With
 Introduction to intelligent network design driven by big data analytics  3

the help of system Verilog language, the developed reusable testbench is used for
verification. The injected inputs to the testbench are randomized with constraints,
such that the design should produce accurate output. To unify the verification lan-
guage there is a dedicated methodology commonly known as Universal Verification
Methodology (UVM); by this, the article is extended to experience the readers also
through the coverage-­based formal verification. For continuous functional verifi-
cation, an intelligent regression model is also developed with the help of ML and
scripting. With this repeated injection of various test cases is possible in order to ver-
ify the functionality. Thus, with the adoption of the presented verification environ-
ment and distinctive approach, one can affirm that the design is ready to be deployed
over the targeted semiconductor chips. As the verification is an unignorable proce-
dure, this can be used to classify the algorithms developed in ML for data clustering,
data encoding and its accurate analysis. More importantly, this chapter allows us to
understand an intelligent verification model for testing the design with regression
run with the corresponding set-­up and the pass/failure analysis steps. This structure
may result in a significant reduction of the simulation time for a VLSI verification
engineer.
Keywords: VLSI, verification, intelligent, ASIC.

Chapter 4: Evaluation of machine learning algorithms on

academic big dataset by using feature selection techniques
Identifying the most accurate methods for forecasting students’ academic achieve-
ment is the focus of this research. Globally, all educational institutions are con-
cerned about student attrition. The goal of all educational institutions is to increase
the student’s retention and graduation rates and this is only possible if at-­risk stu-
dents are identified early. Due to inherent classifier constraints and the incorporation
of fewer student features, most commonly used prediction models are inefficient
and incur. Different data mining algorithms like classification, clustering, regres-
sion, and association rule mining are used to uncover hidden patterns and relevant
information in student performance big datasets in academics. Naïve Bayes, random
forest, decision tree, multilayer perceptron (MLP), decision table (DT), JRip, and
logistic regression (LR) are some of the data mining techniques that can be applied.
A student’s academic performance big dataset comprises many features, none of
which are relevant or play a significant role in the mining process. So, features with
a variance close to 0 are removed from the student’s academic performance big data-
set because they have no impact on the mining process. To determine the influence
of various attributes on the class level, various feature selection (FS) techniques
such as the correlation attribute evaluator (CAE), information gain attribute evalu-
ator (IGAE), and gain ratio attribute evaluator (GRAE) are utilized. In this study,
authors have investigated the performance of various data mining algorithms on
the big dataset, as well as the effectiveness of various FS techniques. In conclu-
sion, each classification algorithm that is built with some FS methods improves the
performance of the classification algorithms in their overall predictive performance.
4  Intelligent network design driven by Big Data analytics

Keywords: data mining, classification, Big Data; feature selection, correla-

tion attribute evaluator, information gain attribute evaluator, gain ratio attribute
evaluator.

Chapter 5: Accurate management and progression of Big Data

analysis

Statistical investigations are concerned with circumstances including arranging,

information assortment, association of data, and scrutiny of compiled data, conver-
sion, and exposure in a clear and chosen procedure. To do so, research techniques
can be distinguished in two different ways: conclusion overviews and statistical sur-
veying. In assessments of public sentiment, the primary objective is to assemble
data about deciding subjects dependent on close-­to-­home meetings. Statistical sur-
veying is directed through the market investigation of a specific item. Assortment,
association, depiction of information, estimation, and translation of coefficients
have a place with descriptive statistics, while investigation and understanding of
information, related to an edge of vulnerability, is the obligation of the inductive or
inferential statistics, additionally called the proportion of vulnerability or techniques
that depend on likelihood hypothesis [9, 10]. The utilization of tables and charts is
visited in statistics. The tables serve to arrange and classify the information, and the
illustrations pass on the data with lucidity and straightforwardness, adding to a goal
pursuing.
Keywords: Big data analysis, scheduling, statistics, machine learning, enor-
mous information, cybercrime, Internet of things (IoT).

Chapter 6: Cram on data recovery and backup cloud computing

techniques

The present digital world technology is evolving at a rapid pace. To store, manage
and protect the digital information, it is necessary to back up and recover the data
with utmost efficiency. As a solution, cloud computing that offers customers a wide
range of services can be used [11, 12]. Storage-­as-­a-­Service (SaaS) is one of the
cloud platform’s services, in which a large volume of digital data is maintained in
the cloud database. Enterprise’s most sensitive data are stored in the cloud, ensuring
that it is secure and accessible at all times and from all locations. At times, informa-
tion may become unavailable due to natural disasters such as windstorms, rainfall,
earthquakes, or any technical fault and accidental deletion. To ensure data security
and availability under such circumstances, it is vital to have a good understanding
of the data backup and recovery strategies. This chapter examines a variety of cloud
computing backup and recovery techniques.
Keywords: cloud computing, data backup, data recovery, advantage,
disadvantage.
 Introduction to intelligent network design driven by big data analytics  5

Chapter 7: An adaptive software defined networking (SDN) for

load balancing in cloud computing
The Internet of Things (IoT) can be perceived as a collection of millions of devices
that are connected among each other and with the Internet as a connectivity back-
bone to acquire and share real-­time data for providing intelligent services. The tre-
mendous rise in the number of devices requires an adequate network infrastructure
to remotely deal with data orchestration. To overcome this issue, a new approach
of infrastructure sharing over the cloud among service providers has transpired,
with the goal of lowering excessive infrastructure deployment costs. The software-­
defined networking (SDN) is a networking architecture that enables network opera-
tors and users to monitor and manage the network devices remotely and more flex-
ibly by using software that runs on external servers. As SDN and cloud integration
improves reliability, scalability, and manageability, this paper combines cloud infra-
structure with SDN. Although SDN-­based cloud networks have numerous advan-
tages as mentioned above, there still exist certain challenges that draw the attention
of researchers like energy efficiency, security, load balancing, and so on. The work
carried out in this chapter is an attempt to address one of the challenging tasks,
namely the load balancing, by developing a new multiple-­controller load-­balancing
strategy. The proposed strategy effectively balances the load even if one or more
super controllers fail. Furthermore, results are simulated and compared under differ-
ent operational environments, both with and without the Modified Bully algorithm.
The comparison results ensure that the introduced technique exhibits better perfor-
mance with metrics such as packet loss, packet transmission ratio, and throughput.
Keywords: SDN, load balancing, IoT, controller, packet loss, transmission
ratio, throughput.

Chapter 8: Emerging security challenges in cloud computing: An

insight
Cloud computing has been evolved as a new computing prototype with the aim of
providing reliability, quality of service and cost-­effective with no location barrier.
More massive databases and applications are relocated to an immense centralized
data center known as the cloud. Cloud computing has enormous benefits of no need
to purchase physical space from a separate vendor instead of using the cloud, but
these benefits have security threats [13, 14]. The resource virtualization, the data
and the machine are physically absent in the cloud; the storage of data in the cloud
causes security issues. An unauthorized person can penetrate through the cloud
security and can data manipulation, data loss or theft might take place. This chapter
has described cloud computing and various security issues and challenges that are
present in different cloud models and cloud environment. It also gives an idea of
different threat management techniques available to encounter security issues and
challenges. The RSA algorithm implementation has been described in detail, and
the Advance Encryption Standard policy, along with its implementation, has also
6  Intelligent network design driven by Big Data analytics

been discussed. For better clarification, several reviews are conducted on the exist-
ing models.
Keywords: security, cloud computing, saas, pass, iaas, cryptography.

Chapter 9: Factors responsible and phases of speaker

recognition system

The method of identifying a speaker based on his or her speech is known as automatic
speaker recognition. Speaker/voice recognition is a biometric sensory device that
recognizes people by their voices. Most speaker recognition systems nowadays are
focused on spectral information, which means they use spectral information derived
from speech signal segments of 10–30 ms in length. However, if the received speech
signal contains some noise, the cepstral-­based system’s output suffers. The primary
goal of the study is to see the various factors responsible for improved performance
of the speaker recognition systems by modeling prosodic features, and phases of
speaker recognition system. Furthermore, in the presence of background noise, the
analysis focused on a text-­independent speaker recognition system.
Keywords: voice recognition, signals, noise, quality.

Chapter 10: IOT-based water quality assessment using fuzzy

logic controller

Water is an essential resource that we use in our daily life. The standard of the water
quality must be observed in real time to make sure that we obtain a secured and clean
supply of water to our residential areas. A water quality-­monitoring and decision-­
making system (WQMDMS) is implemented for this purpose based on Internet of
Things (IoT) and fuzzy logic (FLC) to decide the usage of water (drinking or tap
water) in a common water tank system. The physical and chemical properties of
data are obtained through continuous monitoring of sensors. The work describes in
detail the design of a fuzzy logic controller for a water quality measurement system,
to determine the quality of water by decision-­making, and accordingly, the usage of
water is decided. The WQMDM system measures the physico-­chemical characteris-
tics of water like pH, turbidity, and temperature by the use of corresponding analog
and digital sensors. The values of the parameters obtained are used to detect the
presence of water contaminants and accordingly, the quality of water is determined.
The measurements from the sensor are handled and processed by Esp32 and these
refined values follow the rules determined by the fuzzy inference system. The output
highlights the water quality that is categorized as very poor, poor, average, good.
The usage of the water will be determined by the results obtained using the FLC and
as per the percentage of water quality, the water is decided as drinking water or tap
water.
Keywords: WQMDMS, IoT, controller, fuzzy logic.
 Introduction to intelligent network design driven by big data analytics  7

Chapter 11: Design and analysis of wireless sensor network for

intelligent transportation and industry automation
Wireless sensor networks (WSNs) are dense networks and combination of various
distributed heterogeneous sensor nodes in terms of battery power, data storage and
computational capacity. WSNs are very useful in industry automation-­Internet of
things (IOT), agriculture, economic, commercial and various monitoring applica-
tions in digital world. Energy is limited in wireless devices and in many conditions, it
is not a replaceable or chargeable resource. One of the most characteristics of WSNs
is that they are strongly coupled with their application. This chapter is based on the
design and analysis of WSN for intelligent transportation and industry automation.
Proposed work has been validated in NS2 and compared to other existing protocols.
The simulation results are based on throughput, PDR, and delay parameters.
Keywords: WSN, Wi-­Max, NS2, sensors.

Chapter 12: A review of edge computing in healthcare Internet

of Things: theories, practices, and challenges
The pandemic has forced industries to move immediately their critical workload to
the cloud in order to ensure continuous functioning. As cloud computing expansions
pace and organisations strive for methods to increase their network, agility and stor-
age, edge computing has shown to be the best alternative. The healthcare business
has a long history of collaborating with cutting-­edge information technology, and
the Internet of Things (IoT) is no exception. Researchers are still looking for sub-
stantial methods to collect, view, process, and analyze data that can signify a quanti-
tative revolution in healthcare as devices become more convenient and smaller data
become larger. To provide real-­time analytics, healthcare organisations frequently
deploy cloud technology as the storage layer between system and insight. Edge com-
puting, also known as fog computing, allows computers to perform important analy-
ses without having to go through the time-­consuming cloud storage process [15,
16]. For this form of processing, speed is key, and it may be crucial in constructing
a healthcare IoT that is useful for patient interaction, inpatient treatment, popula-
tion health management and remote monitoring. We present a thorough overview
to highlight the most recent trends in fog computing activities related to the IoT in
healthcare. Other perspectives on the edge computing domain are also offered, such
as styles of application support, techniques and resources [17]. Finally, necessity of
edge computing in era of Covid-­19 pandemic is addressed.
Keywords: edge computing, fog computing, cloud computing, IOT, machine
learning, healthcare.
8  Intelligent network design driven by Big Data analytics

Chapter 13: Image processing for medical images on the basis of

intelligence and bio computing

Intelligence in medical imaging explores how intelligent computing can cre-

ate large number of changes to existing technology in the field of medical image
processing. The book presents various algorithms, techniques and models toward
integrating medical image processing with artificial intelligence and biocomput-
ing. Bioinformatics solutions lead to an effective method for processing the image
data for the purpose of retrieving the information of interest and to collect various
data sources for extracting the knowledge. Moreover, image processing methods
and techniques help the scientists and physicians in medical field for diagnosis
and therapies. It describes evolutionary optimization techniques, support vector
machines (SVMs), fuzzy logic, a Bayesian probabilistic framework, a reinforce-
ment learning-­based multistage image segmentation algorithm, a machine learning
(ML) approach. It discusses how these techniques are used for image classification,
image formation, image visualization, image analysis, image management, and
image enhancement. The term “medical image processing” illustrates the provision
of digital image processing particularly for medicine. Medical imaging intends to
identify internal structures hidden in the human body. It helps to find out abnor-
malities in the body. Digital images can be processed effectively, also evaluated and
utilized in many circumstances concurrently with help of suitable communication
protocols.
Keywords: image processing, machine learning, CNN, deep learning, medical images.

Chapter 14: IoT-based architecture for smart health-care

systems

Internet of Things (IoT) provides a pathway for connecting physical entities with
digital entities using devices and communication technologies. The rapid growth
of IoT in recent days has made a significant influence in many fields. Healthcare
is one of those fields which will be hugely benefited by IoT. IoT can resolve many
challenges faced by patients and doctors in healthcare. Smart health-­care applica-
tions allow the doctor to monitor the patient’s health state without human interven-
tion. Sensors collect and send the data from the patient. Recorded data are stored
in a database that enables medical experts to analyze those data. Any abnormal
change in the status of the patient can be notified to the doctor. This paperwork
aims to study different research works made on IoT-­based health-­care systems that
are implemented using basic development boards. Various hardware parameters of
health-­care systems and sensors used for those parameters are explored. A basic
Arduino-­based health-­care application is proposed using sensors and global system
for mobile communication (GSM) module.
Keywords: IoT, smart healthcare, patient monitoring.
 Introduction to intelligent network design driven by big data analytics  9

Chapter 15: IoT-based heart disease prediction system

In India, around 80 per cent of people who die due to heart disease do not receive
proper treatment. This is a daunting task for doctors as they often do not diagnose
patients properly. The treatment of this disease is very costly. The purpose system
improves the cost-­effective treatment using data mining technology to simplify the
Decision Support System. Most hospitals employee a hospital management system
to manage the care of their patients. Unfortunately, many of these programs do not
use big clinical data to extract important information. As these systems generate a
large quantity of data in various embodiment in spite of the data being rarely looked
upon and remains unused. Therefore, in this process, much effort is required to make
wise decisions. This project helps diagnose a disease using various data mining tech-
niques. Currently, diagnosing a disease involves identifying various symptoms and
features of a disease.
Keywords: heart disease prediction, data mining, deep learning.

Chapter 16: DIAIF: detection of interest flooding using artificial

intelligence-based framework in NDN android
In today’s world, information-­ centric networking (ICN) is a brand-­ new next-­
generation network for distributing multimedia content. ICN focuses on sharing
content across the network rather than obtaining content from a single fixed server.
In-­network caching aids in the dissemination of content from the network, and the
ICN also includes a number of intrusive security mechanisms. Despite the ICN net-
work’s many security measures, several attacks, especially interest flooding attacks
(IFA), continue to wreak havoc on the network’s distribution capability. In order to
address security threats, the literature includes a number of mitigating procedures.
However, legitimate users’ requests are misclassified as an attack in an emergency
circumstance, affecting the network’s QoS. In this chapter, Detection of Interest
Flooding Attack using Artificial Intelligence Framework (DIAIF) is proposed in
ICN. DIAIF seeks to lighten the load on ICN routers by removing the source of the
attack without interfering with legitimate user requests. DIAIF depends on router
feedback to assign a beneficial value (BV) to each piece of content and to block dan-
gerous users based on the BV. The ICN testbed was designed to assess the proposed
DIAIF’s performance in terms of QoS during severe flood scenarios, responding
with malicious content without interfering with genuine user requests, and identify-
ing the source of attack in a communication scenario.
Keywords: Information centric networking, QoS, attacks, flooding attacks,
security threats.
10  Intelligent network design driven by Big Data analytics

Chapter 17: Intelligent and cost-effective mechanism for

monitoring road quality using machine learning
Nowadays, one of the most significant components of road infrastructure is moni-
toring road surface conditions, which leads to better driving conditions and reduces
the chance of a road accident. Traditional road condition monitoring systems are
incapable of gathering real-­time information concerning road conditions. In previ-
ous generations, road surface condition monitoring was done for fixed roadways and
vehicles travelling at a constant pace. Several systems have presented a method for
exploiting the sensors installed in automobiles. However, this method will not assist
in forecasting the precise placement of potholes, speed bumps, or staggered roads.
As a result, smartphone-­based road condition evaluation and navigation are
becoming increasingly popular. We propose exploring several machine learning
techniques to accurately assess road conditions using accelerometer, gyroscope, and
Global Positioning System (GPS) data collected from cellphones. We also recorded
footage of the roadways in order to reduce noise in the data. This two-­pronged
approach to data collection will aid in the exact positioning of potholes, speed
bumps and staggered roads. This method of data collection will aid in the classifica-
tion of road conditions into numerous features such as roads with smooth surfaces,
potholes, speed bumps and staggered highways using machine learning algorithms.
This method of data collection will aid in the classification of road conditions into
numerous features; machine learning algorithms are used to create characteristics
such as smooth roads, potholes, speed breakers, and staggered highways. The user
will receive this information via the map, which will classify the various road con-
ditions. Accelerometers and gyroscope sensors will analyse multiple features from
all three axes of the sensors in order to produce a more precise location of desig-
nated routes. To classify the road conditions, we investigate the performance utilis-
ing support vector machine (SVM), random forest, neural network and deep neural
network. As a result, our findings demonstrate that models trained using a dual data
gathering strategy will produce more accurate outcomes. Data classification will be
substantially more accurate when neural networks are used. The methods described
here can be used on a broader scale to monitor roads for problems that pose a safety
concern to commuters and to give maintenance data to appropriate authorities.
Keywords: SVM, machine learning, GPS data, accelerometer data, android,
road quality, smartphone.

References

[1] Kumar S., Ranjan P., Ramaswami R. ‘EMEEDP: enhanced multi-­hop energy
efficient distributed protocol for heterogeneous wireless sensor network’.
Fifth International Conference on Communication Systems and Network
Technologies; Gwalior, India, IEEE, 2015. pp. 194–200.
[2] Kumar S., Rao A.L.N., Ramaswani R. ‘Energy optimization technique for
distributed localized wireless sensor network’. International Conference
 Introduction to intelligent network design driven by big data analytics  11

on Issues and Challenges in Intelligent Computing Techniques (ICICT);

Ghaziabad, India, IEEE, 2014. pp. 350–55.
[3] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India; 2021. pp. 01–6.
[4] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­ based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems; Springer, Singapore: THDC Institute of Hydropower
Engineering and Technology Tehri Uttarakhand, 2020. pp. 327–36.
[5] Kumar S., Trivedi M.C., Ranjan P., Punhani A. Evolution of Software-­Defined
Networking Foundations for IoT and 5G Mobile Networks. IGI Publisher;
2020.
[6] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[7] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[8] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
personal communications. 2021, vol. 10(3), pp. 1–14.
[9] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road Surface Quality Monitoring
Using Machine Learning Algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[10] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[11] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation
of fault tolerance technique for internet of things (iot)’. Proceedings - 12th
International Conference on Computational Intelligence and Communication
Networks; Bhimtal, India, IEEE, 2020. pp. 154–59.
[12] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of Indian roads’. Proceedings of the
Confluence - 10th International Conference on Cloud Computing; Noida,
India; 2020.
[13] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 4th International Conference on Information Systems and
Computer Networks; Mathura, India, 2019. pp. 250–55.
12  Intelligent network design driven by Big Data analytics

[14] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India, IEEE, 2021. pp. 01–06.
[15] Haidar M., Kumar S. ‘Smart healthcare system for biomedical and health
care applications using aadhaar and blockchain’. 2021 5th International
Conference on Information Systems and Computer Networks, ISCON 2021;
Mathura, India, IEEE, 2022. pp. 1–5.
[16] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic Torus
Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering. 2019, vol. 8(6), pp. 2278–3075.
[17] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
 Chapter 2
Role of automation, Big Data, AI, ML IBN, and
cloud computing in intelligent networks
Sunil Kumar 1 and Priya Ranjan 2

There are more smart devices in our world today than individuals. A growing num-
ber of people are linked to the Internet 24 hours a day, in some form or another.
A growing number of people own three, four, or more smart devices and rely on
them. Smartphones, fitness and health trackers, e-­readers, and tablets are just a few
examples. It is forecast that on average there will be 3.4 smart devices or connections
for every person on earth. The Internet of Things (IoT) is relevant to many indus-
tries. IoT systems contribute to the environmental controls, retail, transportation,
healthcare, and agriculture industries among many others. According to Statista, the
number of IoT devices that are in use across all relevant industries is forecast to
grow to more than 8 billion by 2030. As for consumers, important growth areas are
the Internet and digital media devices, which include smartphones. This area is also
predicted to grow to more than 8 billion by 2030. Other applications with more than
1 million connected devices are connected and autonomous vehicles, IT infrastruc-
ture, asset management, and electric utility smart grid.
All of this is made possible through intelligent networks. The planet is rap-
idly becoming covered in networks that allow digital devices to communicate and
interconnect. Consider the network mesh as a digital skin that surrounds the earth.
Mobile devices, electronic sensors, electronic measuring equipment, medical gad-
gets, and gauges can all link with this digital skin. They keep track of, communicate
with, analyze, and, in some situations, automatically adjust to the data collected and
transmitted.

2.1 Evolution of networks: everything is connected

Hundreds of trillions of gigabytes of data are generated by 30 billion things. How

can they collaborate to help us make better decisions and improve our lives and

1
Department of Computer Science & Engineering, Amity University, Uttar Pradesh, India
2
Department of Electronics & Communication Engineering, SRM University, Andhra Pradesh, India
14  Intelligent network design driven by Big Data analytics

businesses? The networks that we utilize daily enable these relationships [1]. The
Internet and the digitized world are built on the foundation of intelligent networks.
The methods we use to communicate are constantly changing. Breakthroughs in
wireless and digital technology have substantially extended the reach of our com-
munications, which were formerly limited by wires and plugs [2]. Through their
connectivity to the Internet, networks in businesses and large organizations can sup-
ply products and services to clients. Networks can also be used on a larger scale to
consolidate, store, and offer access to data on network servers [3].
The Internet is the world’s largest network, and it serves as the “electronic skin”
that protects the globe. The Internet is a collection of private and public networks
that are linked together. The Internet is accessible to businesses, small office net-
works, and residential networks. The data collected, retained, and analyzed by sen-
sors benefit a wide range of companies. Businesses now have a better understanding
of the things they sell and who is buying them. They can streamline production and
target their marketing and advertising to specific areas or audiences using this type
of data, encouraging the development of new business opportunities, and marketing
ideas [4].

2.1.1  Intelligent devices

The Internet of Things (IoT) is the Internet-­connected network of millions of smart
devices and sensors. Many organizations can use and evaluate the data collected and
shared by these linked devices and sensors. Businesses, cities, governments, hos-
pitals, and individuals are among these organizations. The IoT has been made pos-
sible in part by the availability of low-­cost CPUs and wireless networks. Inanimate
objects such as doorknobs and light bulbs can now be fitted with an intelligent sen-
sor that collects and transmits data to a network [5, 6].
In 2025, researchers predict that 38.6 billion IoT devices will be connected to
the Internet, with 50 billion by 2030. Every month, 190 million new linked devices
are added! Computers, smartphones, tablets, and smart TVs will account for about
a third of all connected devices. Sensors, actuators, and newly designed intelligent
gadgets that monitor, regulate, analyze, and optimize our reality will make up the
remaining two-­thirds. Smart doorbells, garage doors, thermostats, sports wearables,
pacemakers, traffic lights, parking spots, and other intelligent connected sensors are
just a few examples. Only human imagination limits the number of potential objects
that could become sentient sensors [7].

2.1.2  Intelligent devices connection with networks

The obtained data must be stored and shared; hence an intelligent device must be
connected to a network. A wired ethernet connection or a wireless link to a controller
is required. Low-­power alternatives such as Bluetooth LE, Zigbee, or LoRa are more
practical than wireless ethernet. Sensor data are collected by controllers, while net-
work or Internet access is provided by controllers. Controllers may be able to make
judgments right away or send data to a more powerful computer for study. This more
 Role of automation, Big Data, AI, ML IBN, and cloud computing  15

powerful computer may be on the same LAN as the controller, or it could only be
reached via the Internet [8, 9].
Actuators are frequently used in conjunction with intelligent devices. Actuators
are devices that accept electrical input and convert it into physical action. For exam-
ple, if a smart device senses excessive heat in a room, the temperature reading is sent
to the microcontroller. The data can be sent from the microcontroller to an actuator,
which will then switch on the air conditioner. The majority of new devices such
as fitness wearables, implanted pacemakers, air meters in a mine shaft, and water
meters in a farm field all require wireless connectivity. Because many sensors are
“out in the field” and are powered by batteries or solar panels, consideration must
be given to power consumption. Low-­powered connection options must be used to
optimize and extend the availability of the sensor [10–12].

2.2 Huge volume of data generation by intelligent devices

Data are collected from a variety of sources, including individuals, photographs,

text, sensors, and online sites. Cell phones, PCs, kiosks, tablets, routers, switches,
firewalls, and cash registers all contribute data. The volume of data created by intel-
ligent gadgets has recently increased dramatically. Intelligent gadgets are currently
found in an increasing number of places and items. Security cameras, traffic lights,
intelligent cars, thermometers, and even grape plants are among them. Big Data
refer to a large amount of data, but how large is a large amount of data? No one can
say with certainty whether data from a company are considered “Big Data.” The fol-
lowing are three indicators that suggest a company is dealing with Big Data:

• They have a significant amount of data that require more storage space as time
goes on (volume).
• They are dealing with an ever-­increasing amount of data (velocity).
• They have information in a number of formats (variety).

How much data do intelligent devices collect? Here are some estimated exam-
ples. For comparison, assume that the average MP3 song is about 3 megabytes.

• Intelligent devices in one smart connected home can produce as much as 1 giga-
byte (GB) of information a week, or the equivalent of 333 MP3 songs.
• Intelligent devices in one autonomous car can generate 4,000 gigabits (Gb) of
data per day. That is, 500 gigabytes (GB) of data, which is the equivalent of
about 167,000 MP3 songs.
• Safety intelligent devices in mining operations can generate up to 2.4 terabits
(TB) of data every minute. That is, 300 GB or about 100,000 MP3 songs.
• An Airbus A380 Engine generates 1 petabyte (PB) of data on a flight from
London to Singapore. That is, 1 million GB or about 334 million MP3 songs.
16  Intelligent network design driven by Big Data analytics

2.2.1  Issues and challenges of Big Data Analytics

The World Economic Forum predicts that the amount of data generated daily will be
463 exabytes (EB) globally. One EB is equal to 1 billion gigabytes! To put this into
context, according to Statista, every minute of every day:

• 500 hours of YouTube videos uploaded daily by different devices.

• 69 million instant messages shared by different devices.
• Over 347,000 GB of Netflix video uploaded and shared daily.
• 198 million emails sent on a daily basis.
• Over 60,000 Instagram images shared by devices daily.

The rapid growth of data can be an advantage or an obstacle when it comes to

achieving business goals. To be successful, enterprises must be able to easily access
and manage their data assets. With this enormous amount of data being constantly
created, traditional technologies and data warehouses cannot keep up with stor-
age needs. Even with cloud storage services from firms such as Amazon, Google,
Microsoft, and others, the security of stored data becomes a significant issue. To
prevent data loss, Big Data systems must be safe, have high fault tolerance, and use
replication. It is not only about storing data; when it comes to Big Data storage, it is
also about managing and securing it [13–15].

2.2.2 Storage of Big Data

Big Data are typically stored on multiple servers, usually housed within data centers.
The information is generally spread and replicated on numerous servers in different
data centers for security, accessibility, and redundancy [16].

2.2.2.1 Edge computing
Edge computing (Figure 2.1) is an architecture that utilizes end-­user clients or
devices at the edge of the network to do a substantial amount of the preprocessing
and storage required by an organization. Edge computing was designed to keep the
data closer to the data source for preprocessing. Intelligent device data, in particu-
lar, can be preprocessed closer to where it was collected. The information gained
from that preprocessed analysis can be fed back into companies’ systems to mod-
ify processes if required. Because the sensor data are preprocessed by end devices
within the company system, communications to and from the servers and devices are
quicker. This requires less bandwidth than constantly sending raw data to the cloud
[17, 18]. After the data have been preprocessed, it is often shipped off for long-­term
storage, backup, or deeper analysis within the cloud [19].

2.2.2.2 Cloud and computing

The cloud is a collection of data centers or group of connected servers. Access
to software, storage, and services available on the servers is obtained through the
Internet via a browser interface [20, 21]. Cloud services are provided by many large
 Role of automation, Big Data, AI, ML IBN, and cloud computing  17

Figure 2.1    Edge computing

companies such as Google, Microsoft, and Apple. Cloud storage services are pro-
vided by different vendors such as Google Drive, Apple iCloud, Microsoft OneDrive,
and Dropbox. From an individual’s perspective, using the cloud services allows you:

• to access various programs instead of downloading them onto your local device
• to save all of your data, such as images, music, movies, and emails, freeing up
local hard disc space
• to be able to access your data and applications from any location, at any time,
and on any device

One of the disadvantages of using the cloud is that your data could fall into the
wrong hands. Your data are at the mercy of the security robustness of your chosen
cloud provider. From the perspective of an enterprise, cloud services and computing
support a variety of data management issues.

2.2.2.3 Distributed processing
From a data management perspective, analytics were simple when only humans cre-
ated data. The amount of data was manageable and relatively easy to sift through.
However, with the explosion of business automation systems and the exponential
growth of web applications and machine-­generated data, analytics is becoming
increasingly more difficult to manage. In fact, 90 percent of the data on the planet
now were created in the last 2 years. Exponential growth is characterized by an
increase in volume over a short period of time. This high volume of data is difficult
18  Intelligent network design driven by Big Data analytics

to process and analyze within a reasonable amount of time. Rather than large data-
bases being processed by big and powerful mainframe computers and stored in giant
disk arrays (vertical scaling), distributed data processing takes the large volume of
data and breaks it into smaller pieces. These smaller data volumes are dispersed over
multiple sites to be processed by a large number of computers with less powerful
CPUs. Each computer in the distributed architecture examines its own piece of the
Big Data puzzle (horizontal scaling).
Most distributed file systems are designed to be invisible to client programs.
The distributed file system locates files and moves data, but the users have no way
of knowing that the files are distributed among many different servers or nodes. The
users access these files as if they were local to their own computers. All users see
the same view of the file system and are able to access data concurrently with other
users. Hadoop was built to deal with these massive amounts of data. The Hadoop
project began with two components: MapReduce is a distributed, fault-­tolerant file
system based on Hadoop Distributed File System, which is a distributed way to
process data. Hadoop has now evolved into a very comprehensive ecosystem of
software for Big Data management. Hadoop is open-­source software enabling the
distributed processing of large data sets that can be terabytes in size and that are
stored in clusters of computers. Hadoop is designed to scale up from single servers
to thousands of machines where they offer computation and storage. To make it
more efficient, Hadoop can be installed and run on many virtual machines (VMs).
These VMs can all work together in parallel to process and store the data.
Hadoop uses scalability and fault tolerance as two important features:

• Scalability: With Hadoop, cluster size can easily scale from a 5-­node cluster to
a 1,000-­node cluster without excessively increasing the administrative burden.
• Fault tolerance: Hadoop creates many replicated files automatically and pro-
vides backup to ensure that data will not be lost. If a disk, node, or a whole rack
fails, the data are safe.

2.3 Need of data analysis by business

Every organization must become more efficient and innovative to stay competitive
and relevant in the digitized world. The IoT is an integral part of achieving that effi-
ciency and innovation. Many companies want to collect and analyze large amounts
of new product usage data in order to acquire valuable insights. Businesses can use
data analytics to better evaluate the impact of their products and services, alter their
processes and aims, and give better products to their consumers faster. The ability to
gain new insights from their data brings value to the business.
To businesses, data are the new oil. Like crude oil, it is valuable, but if it is
unrefined, it cannot be easily used. Crude oil has to be changed to gasoline, plastic,
chemicals, and other substances to create a valuable product. It is the same with
data. Data must be broken down and analyzed for it to have value. Transactional
 Role of automation, Big Data, AI, ML IBN, and cloud computing  19

and analytical data are the two main types of processed data that provide value.
As events occur, transactional data are recorded and processed. Daily sales reports
and production schedules are analyzed using transactional data to decide how much
inventory to keep on hand. Managerial analysis such as assessing whether the com-
pany should establish a new manufacturing plant or hire more salespeople is aided
by analytical data.

2.3.1  Sources of information

The source of data in the large data sets is varied. Apart from sensor data, other
data originate from anything that has been scanned, entered, and released into the
Internet from sources such as:

• Social media sites: Facebook, YouTube, WhatsApp, WeChat, TikTok, and

Instagram
• HTTP, content search, net surfing data
• Archived data
• Metadata that is attached to emails, transmitted documents, and pictures
• Medical forms, insurance forms, and tax forms
• Genomics research using DNA

Even if data are considered structured, different applications create files in dif-
ferent formats that are not necessarily compatible with one another. Structured data
may need to be manipulated into a common format such as CSV. Comma-­separated
value (CSV) files are a type of plaintext file that uses commas to separate columns
in a table of data and the carriage return character to separate rows. Each row is a
record. Although they are commonly used for importing and exporting in traditional
databases and spreadsheets, there is no specific standard. Data formatting techniques
such as JSON and XML are also plaintext file types that use a standard way of
representing data records. These file formats are compatible with a wide range of
applications.
Unstructured data require different tools to prepare data for processing or analy-
sis. The following are two examples:
Web pages are created to provide data to humans, not machines. “Web scrap-
ing” tools automatically extract data from HTML pages. This is similar to a Web
Crawler or spider of a search engine. It explores the web to extract data and cre-
ates a database to respond to the search queries. The web scraping software may
use Hypertext Transfer Protocol or a web browser to access the World Wide Web.
Typically, web scraping is an automated process that uses a bot or web crawler to
do data mining. Specific data are gathered and copied from the web to a database or
spreadsheet. The data can then be easily analyzed.
Many large web service providers such as Facebook provide standardized
interfaces to collect the data automatically using application programming inter-
faces (APIs). The most common approach is to use RESTful APIs. RESTful APIs
use HTTP as the communication protocol and JSON structure to encode the data.
20  Intelligent network design driven by Big Data analytics

Internet websites such as Google and Twitter gather large amounts of static and
time-­series data. Knowledge of the APIs for these sites allows data analysts and
engineers to access the large amount of data that are constantly being generated on
the Internet.

2.3.2 Data visualization
The mined data must be analyzed by intelligent tools and techniques and should be
presented to managers and decision-­makers to be of value with minimum errors.
There are many different visualizations that can be used to present the value in the
data. Determining the best chart to use will vary based on the following:

• Number of variables to show

• Number of data points in each variable
• Is the data representing a timeline?
• Items that require comparisons
• Some of the most popular chart types are line, column, bar, pie, and scatter.

2.3.3 Analyzing Big Data for effective use of business

Big Data is just that – BIG! It is most helpful to analyze it to get value out of it.
Analyzing big data typically requires tools and applications created for this purpose.
These analysis tools have been designed to provide businesses with detailed infor-
mation, patterns, and valuable insights.
Before beginning any analysis, it is critical to know what problem the business
is trying to solve or what information the business is looking for. Are they interested
in customer behavior in specific states, energy consumption patterns in different city
quadrants, or the number of Facebook “likes” based on age? There are many desktop
Big Data Analytics tools that businesses could select such as Knime, OpenRefine,
Orange, and RapidMiner. Cloud-­based analytics tools include Google’s Big Query,
IBM Cognos Analytics, TIBCO Spotfire, and Board.

2.3.4 Intelligent devices thinking intelligently

Can things think? Can a device learn from its environment? In this context, there
are many definitions of “think.” One possible definition is the ability to connect a
series of related pieces of information together and then use them to alter a course
of action. For example, when we are young, we have no concept that fire is hot and
that placing our hand in the fire will cause pain. A fire may appear visually pleasing
and actually encourage one to try and touch the flames. We quickly learn that the fire
can cause injury. We then start to associate the image of the fire with the pain it can
cause. From this point onward, we start to think about the results of touching the fire
and base our actions on this acquired information.
Many products now include smart technology that allows them to change their
behavior in particular situations. This can be as basic as a smart appliance that
reduces its power consumption during peak demand periods or as complex as a
 Role of automation, Big Data, AI, ML IBN, and cloud computing  21

self-­driving car. When a gadget makes a choice or takes a course of action based
on information from the outside world, it is referred to as a smart device. The word
“smart” now appears in the titles of many of the devices with which we interact.
This suggests that the device can change its behavior in response to its surroundings.

2.4 Artificial intelligence and machine learning in networking

Machine intelligence is referred to as artificial intelligence (AI). Natural intel-

ligence, on the other hand, is the intelligence demonstrated by living beings. AI
employs intelligent agents who are able to assess their surroundings and make
actions that increase the likelihood of achieving a given goal or purpose. AI refers to
systems that replicate cognitive functions such as learning and problem-­solving that
are traditionally associated with human minds. Autonomous cars, intelligent routing
in content delivery networks, strategic game playing, and military simulations are
some of the jobs that are now considered to require some level of AI.
Many of the tasks that once required AI have become ordinary as technology
advances. Many of these tasks have shifted from AI to machine learning (ML). ML
is a branch of AI that employs statistical techniques to enable computers to “learn”
from their surroundings. This allows computers to improve at a task without having
to be properly trained for it. This is especially useful when designing and program-
ming specific algorithms is difficult or infeasible. One objective of learning is to be
able to generalize based on experience. For machines, this involves the ability to
perform accurately on new, previously unseen tasks after gaining experience with
a learning data set. The training data set must come from data that is representative
of the larger data pool. This data pool enables the machine to build a general model
about this data, which would help it make accurate predictions.

2.4.1  Role of ML in networks

One of the features of the IoT is that it enables the collection of huge pools of data
that can “teach” programs how to respond to certain conditions. Some of the more
common uses of ML technology include the following.

2.4.1.1 Speech recognition
Many different companies now offer digital assistants which allow you to use speech
to communicate with a computer system. Apple, Microsoft, Google, and Amazon all
offer this service. These companies not only allow commands to be given verbally
but also offer speech-­to-­text capabilities.

2.4.1.2 Product recommendation
Systems build up a customer profile and recommend products or services based on
previous patterns. Users of Amazon and eBay receive recommendations on prod-
ucts. Organizations such as LinkedIn, Facebook, and GooglePlus recommend users
you may wish to connect with.
22  Intelligent network design driven by Big Data analytics

2.4.1.3 Credit card fraud detection

A profile is constructed about the purchasing patterns of a client. Any deviation
from these patterns triggers an alert and the system automatically takes action. This
action ranges from denying the transaction to notifying the authorities. Some of the
events that are detected and could indicate a fraudulent transaction include purchas-
ing products not normally purchased, purchases in a different geographical area,
rapidly purchasing many different products, and purchasing large-­ticket items.

2.4.1.4 Facial recognition
Security cameras are everywhere, from stores and streets to airports and transporta-
tion hubs. These cameras continually scan the crowds, normally watching for dan-
gerous or illegal activities, but they can also be used to identify and track individu-
als. The system builds a pattern of specific facial features and then watches for a
match to these facial patterns triggering some action.

2.4.1.5 Anomaly detection
In manufacturing, mining, transportation, and other areas, ML can be used to learn
about normal operating conditions in mechanical systems. This makes it possible
to detect unusual operating conditions that could signal that something is ready to
break. Over time, ML not only detects anomalies but also suggests what the most
likely cause of the anomaly may be. Anomaly detection is commonly used in pre-
ventive maintenance but can also be used to detect other unusual conditions that
could signal safety or security issues in other types of systems such as data networks.

2.5 Intent-based networking

For a business to survive, it must be agile and respond quickly to the needs and
demands of its customers. Businesses are increasingly dependent on their digital
resources to meet customer demands, so the underlying IT network must also be
responsive enough to quickly adapt to these requirements. This normally involves
adjustments to many systems and processes. These adjustments may include
changes to security policies and procedures, business services and applications, and
operational policies. With traditional networks, many different components must
be manually adjusted to meet ever-­changing business requirements. This requires
different technicians and engineers to ensure that the systems are changed to allow
them to work together to accomplish their goals. This sometimes results in errors
and delays and often in suboptimal network performance.
In order to be nimble, responsive, and business-­relevant, the new business network
must seamlessly and securely incorporate IoT devices, cloud-­based services, and remote
offices. The network must also protect these new digital activities from the constantly
evolving threat landscape. To meet this demand, the IT industry has begun to develop
a systematic strategy to link infrastructure management to business goals. Intent-­based
networking is the name given to this method. The diagram depicts the general concept of
 Role of automation, Big Data, AI, ML IBN, and cloud computing  23

intent-­based networking. Business needs are automatically and continuously converted

into IT infrastructure implementation under this new paradigm.

2.6 Role of programming

It is usual for programmers to write the first draft of a program in a programming lan-
guage that they are unfamiliar with. These language-­independent programs, which are
typically referred to as algorithms, are centered on logic rather than syntax. A flowchart
is a visual representation of an algorithm. System software and application software are
the two most popular types of computer software. Application software programs are
designed to perform a certain task or set of tasks. Cisco Packet Tracer, for example, is
a network simulation application that allows users to model complicated networks and
ask “what if” network behavior questions. System software connects the hardware of
the computer to the application program. The system software is responsible for control-
ling the computer hardware and allowing application programs to run. Linux, Apple
OSX, and Microsoft Windows are all instances of system software. A programming
language is used to construct both system and application software. A programming
language is a set of rules for writing programs that send instructions to computer hard-
ware. Algorithms are self-­contained, step-­by-­step sets of operations to be done, and these
programs implement them. Some programming languages compile their code into a col-
lection of machine instructions. C++ is an example of a computer language that is com-
piled. Others do not compile these instructions into machine code before interpreting
them. An interpreted programming language such as Python is an example. The process
of creating a program can begin once the programming language has been decided and
the process has been diagrammed in a flowchart. Program architectures are similar in
most computer languages.

2.6.1  Basic programming using Blockly

Blockly is a visual programming tool created to help beginners understand the con-
cepts of programming. By using a number of block types, Blockly allows a user
to create a program without entering any lines of code. Blockly implements visual
programming by assigning different programming structures to colored blocks. The
blocks also contain slots and spaces to allow programmers to enter values required
by the structure. Programmers can connect programming structures together by
dragging and attaching the appropriate blocks. Programming structures such as con-
ditionals, loops, and variables are all available for use.
Creating a new variable in Blockly is a simple matter of dragging the variable
block onto the workspace and filling in the value slot. It is also possible to change
the contents of a variable as the program is being executed. Blockly also supports
functions. Similar to the variables, Blockly has specific blocks to represent func-
tions. Also similar to variables, programmers simply select and drag function blocks
to the workspace and fill in the required slots. The blocks are available such as an IF-­
THEN block, a WHILE block, and a FOR block. There are also blocks specifically
24  Intelligent network design driven by Big Data analytics

for sensors and actuators. Blockly can translate its block-­based code into Python or
JavaScript. This is very useful to beginner programmers.

2.6.2 Blockly games
Google offers a number of free and open-­source instructional games to aid program-
ming learning. Blockly Games is the name of the game series.
Visit https://blockly.games to discover more about Blockly Games or to give it
a try.
To assist you to get started, there are several stages to accomplish. Blockly may
appear to be a toy, but it is a fantastic tool for honing your logical reasoning skills,
which is one of the fundamentals of computer programming. The first part of this
session covered how to use basic programming to support IoT devices. Flowcharts
are diagrams that show how processes work. System software and application soft-
ware are the two most popular types of computer software. Application software is
designed to complete a certain task.
Variables in programming can be divided into two groups:

• Local variables: Variables that are contained within a program, function, or

process.
• Global variables: These are variables that are in scope during the execution of
the program. Any section of the program can retrieve them.

IF-­THEN, FOR Loops, and WHILE Loops are the most frequent logic constructs.
Blockly is a visual programming tool designed to assist novices in learning pro-
gramming fundamentals. Blockly uses colored blocks to implement visual program-
ming by assigning different programming structures to them.
Python is a widely used programming language that is intended to be simple to
learn and write. Python is an interpreted language, so parsing and executing Python
code necessitates the use of an interpreter.

2.7 Role of technology to design a model

Technology helps to have design, electrical, and physical/mechanical skills while

prototyping in the IoT (working with your hands to put things together). Skills in
networking and programming are also incredibly useful. IoT developers also ben-
efit from understanding cloud computing as well as network and device security.
Because the IoT is still in its early stages, there are still unknown tasks to be dis-
covered. This is an excellent time to create anything that is connected to the IoT.
Because the IoT connects people, processes, data, and things, the number of inven-
tions that the IoT can help generate and subsequently include is limitless.

• Is totally working but not without flaws.

• Is a fully functional version of the product.
• Is used to assess product performance and make improvements.
 Role of automation, Big Data, AI, ML IBN, and cloud computing  25

• Has a fully functional interior and exterior.

• The cost of production may be high.
• Is frequently utilized as a technology demonstration in the IoT.

How do you create a prototype? There are several options for getting started.
The Google Glass was created by a team at Google using the “Rapid Prototyping
Method.” To see a video regarding Google’s approach to prototyping, look up
“google glass quick prototype TED talk” on the Internet. Google, of course, has
plenty of cash to pay for the people and materials involved in prototyping. Most of
us will require financial assistance to bring our ideas from our thoughts to a proto-
type. There is crowdfunding for us. Kickstarter, Indiegogo, and Crowdfunder are
just three of the several online crowdfunding platforms available. Look for “Pebble
Time Kickstarter Video” on the Internet.
Of course, the Internet is a good place to start. People have been exchanging
ideas for millennia, but the Internet has taken it to a whole new level. People who
have never met in person are able to interact and work together now. There are a
number of websites where you may connect with other creators. Maker Media is a
global platform that brings together makers to share projects and ideas. Makers can
also use the platform to find and purchase products for their projects. Make a search
for Makezine on the Internet for further information.

2.7.1  Electronic toolkits

Computer programs cannot function without the use of a computer. While you
can write programs for practically any computer, there are several platforms that
are specifically meant for beginners. Some of the most popular platforms are
listed here. Arduino is an open-­source physical computing platform that consists
of a simple microcontroller board and a software development environment. You
may create interactive objects that operate lights, motors, and other physical
things using input from a variety of switches or sensors. Go to http://arduino.​
cc to learn more about Arduino. While the Arduino is not designed to be used
as a computer, its low power consumption allows it to effectively control other
devices.
The Raspberry Pi is a low-­cost, credit-­card sized computer that connects to a
computer monitor or television. A regular keyboard and mouse are used to oper-
ate it. It can do all of the functions of a computer, including accessing the Internet
and watching high-­definition videos, as well as spreadsheets, word processing, and
game play. Go visit http://www.raspberrypi.org to learn more about Raspberry Pi.
In terms of size, power consumption, and application, the Beaglebone is quite simi-
lar to the Raspberry Pi. Because the Beaglebone has more computing capability
than the Raspberry Pi, it is a superior choice for applications that demand more
processing. Go to http://beagleboard.org to learn more about Beaglebone. Finally,
Adafruit is a great place to start with IoT projects.
26  Intelligent network design driven by Big Data analytics

2.7.2 Programming resources
Programming is essential in the IoT. When designing an IoT solution, bespoke code
is really handy. Blockly and Python have already been introduced to you. There are
plenty of other free tools available to assist you in honing your programming skills.
MIT OpenCourseWare (OCW) is a web-­based repository of nearly all MIT
course materials. OCW is an excellent site to learn about computer programming
for free because it is open to the entire world. http://ocw.mit.edu/courses/intro-pro-
gramming/ contains OCW programming-­related courses.

• Another wonderful resource is Code Academy. It focuses on interactivity to aid

learning.

See? I told you! It does not have to be tough to learn basic programming. It is
possible to have a good time! You now have some very powerful starting tools after
creating a process flowchart and using Blockly and Python. What do you think you
could make for the IoT? How could a small prototype help you get started? It can
be entertaining, such as programming a remote-­controlled toy to play with your cat
while you are away. Programming a heat sensor for a newborn’s bed, for example,
may be lifesaving. I am willing to wager that once you have had some experience
prototyping in the IoT, you will begin to see things differently.

2.8 Relation of AI, ML, and IBN

Aim-­based networking uses automation, AI, and ML to regulate the function of a

network in order to achieve a certain goal or intent. With intent-­based networking,
the IT team can explain exactly what they want the network to do in plain English,
and the network will make it happen. The network can convert the intent into poli-
cies and then automate the deployment of the necessary configurations across the
network.
The intent-­based network employs AI and ML to verify that any services deliv-
ered satisfy the specified service level. The intent-­based network can send alerts and
make ideas for improvement if they do not achieve the service level. The intent-­
based network may be able to automatically alter the network to meet the service
levels in some instances.

2.9 Business challenges and opportunities

The IoT has a lot of advantages, but it also has a lot of drawbacks. We are now faced
with an ever-­expanding collection of new technologies that we must learn since the
IoT is a transformational technology. The IoT is transforming our lives in every way.
This is not the first time we have seen a technical advancement with such ram-
ifications. Farm mechanization enhanced the productivity of accessible farmland
and triggered a population shift from rural to urban areas. The invention of the
 Role of automation, Big Data, AI, ML IBN, and cloud computing  27

vehicle allowed for improved workforce mobility and recreational pursuits. Many
mundane operations could be automated with greater precision and efficiency
thanks to the personal computer. On a worldwide scale, the Internet began to
break down geographical barriers and enhance equality among people. These are
only a few examples of transformative innovations that have occurred in recent
years. Each of these technologies brought significant changes to an established
culture and was initially received with fear and apprehension. The underlying
benefits were apparent when the first dread of the unknown was overcome and the
technology was accepted. Each perceived problem brings with it a slew of new
possibilities.
Can you imagine how different your life would be if you did not have a car, a
computer, or Internet access?

2.9.1  The evolving job market

The IoT is transforming the work market. Jobs that are built to embrace this new envi-
ronment and all it has to offer are replacing traditional jobs. Opportunities in IT may be
limited to edge computing, the development of new processes, or a specialty in a field
that has yet to be realized. In the following domains, these jobs represent abilities from
a variety of fields, including computer science, computer engineering (a combination of
computer science and electrical engineering), and software engineering:

• Collaboration
• Enterprise Networks
• Data Center and Virtualization
• AI
• Application Development
• IoT Program Developer
• IoT Security Specialist

Not all of the jobs created by the IoT are IT-­related. The IoT should be viewed
as an enabling technology with applications in all industries and elements of our
daily life. Within its sphere, the IoT has spawned a plethora of work opportunities.
These positions are available across the design, development, and implementation
of the IoT. There are a few major categories that describe the various career options
available in today’s digital world:

• Enablers: These positions are responsible for developing and implementing the
underlying technology.
• Engagers: These professionals plan, develop, integrate, and provide IoT ser-
vices to customers.
• Enhancers: These positions create their own value-­added services in addition to
Engagers’ services, which are unique to the IoT.
28  Intelligent network design driven by Big Data analytics

The IoT is also creating a need for a new type of IT expert. These are people
who have the knowledge and abilities needed to create new IoT-­enabled goods and
analyze the data they collect.
A workforce with expertise in both information science and software or com-
puter engineering is required. With addition, in the IoT, operational and information
technologies are merging. People must interact and learn from one another in order
to comprehend the things, networks, and procedures that exploit the IoT’s boundless
potential.
We must stay current in order to harness the full potential of what the IoT has
to offer, given the ever-­changing landscape of the digitized world. As new technolo-
gies emerge, the work sector will continue to offer additional opportunities. At the
same time, the skill sets required for these positions will evolve, necessitating the
need for lifelong learning.

2.10 Security

Many companies’ data have been accessed by hackers over the years. The outcome
has been the leaking of millions of users’ data on the Internet, which has had a
tremendous impact. Login passwords and other personal data linked to more than
one million Yahoo and Gmail accounts are purportedly being sold on the dark web
marketplace, according to recent reporting. Usernames, emails, and unencrypted
passwords are purportedly included in the online accounts for sale on the Dark Web.
The accounts are considered to be the result of numerous big cyberattacks rather
than a single data leak.
In July 2017, hackers broke into Equifax (EFX), one of the largest credit
bureaus, and stole the personal information of 145 million people. Because of the
amount of sensitive information revealed, including Social Security numbers, it
was deemed one of the greatest data breaches of all time. Two months later, the
business revealed the hack. Because the stolen data could be exploited for identity
fraud, it could have a long-­term impact. In 2018, the company’s food and nutri-
tion app, MyFitnessPal, was hacked, affecting an estimated 150 million subscribers.
According to the inquiry, usernames, email addresses, and hashed passwords may
have been compromised.

2.10.1  Challenges to secure device and networks

IoT devices are developed with the necessary network connectivity capabilities but
often do not implement strong network security. Network security is a critical fac-
tor when deploying IoT devices. Methods must be taken to ensure the authenticity,
integrity, and security of the data, the path from the sensor to the collector, and the
connectivity to the device.
The number of networked sensors and smart devices is rapidly expanding, pos-
ing a greater risk of assault. Sensors and smart devices are often small devices with
a variety of operating systems, CPU kinds, and memory capacities. Many of these
items will be low-­cost, single-­purpose gadgets with basic network connectivity.
 Role of automation, Big Data, AI, ML IBN, and cloud computing  29

Some IoT devices that are connected to the Internet can interact with the physical
environment. They have found their way into gadgets, automobiles, our bodies, and
our houses. Sensors could collect information from the refrigerator or heating sys-
tem. They could also be found mounted to tree trunks or in city lampposts. Physical
security is difficult or impossible to achieve in many nontraditional locations.
Sensor-­enabled IoT devices may be positioned in inaccessible or remote areas,
making human intervention or configuration nearly impossible. The devices are fre-
quently built to last far longer than ordinary high-­tech equipment. Some IoT devices
are purposely created without the capacity to be upgraded, or they may be placed
in locations where reconfiguring or upgrading is difficult or impossible. Every day,
new vulnerabilities are discovered. If a device cannot be upgraded, the vulnerability
will remain for the duration of its life. If a gadget may be upgraded, the average con-
sumer may not have a technical background; hence the upgrading procedure should
be automated or simple enough for a layperson to complete.

2.10.1.1 Wi-Fi security
Because they are simple to set up and operate, wireless networks are popular in all
types and sizes of enterprises. The organization must provide a wireless experience
that is both mobile and secure for both employees and visitors. Hackers within range
can access and infiltrate a wireless network if it is not properly secured.

2.10.1.2 End-user devices should be protected

The source or destination of data delivered via the network is referred to as an end
device. Any user device, such as computers, tablets, or phones, is considered an end
device. Servers and printers are examples of equipment that users can access or send
data to.
The following are some of the most frequent methods for securing end devices:

i. Activate the firewall

To prevent hackers from accessing your personal or commercial

data, turn on and update your firewall, whether it is a software fire-
wall or a hardware firewall on your router.

ii. Manage your browser and operating system

Hackers are continually looking for methods to exploit flaws in

your operating systems and web browsers. Set the security settings
on your computer and browser to medium or higher to secure your
machine and data. Regularly download and install the newest soft-
ware patches and security updates from the vendors as well as your
computer’s operating system and web browsers.
30  Intelligent network design driven by Big Data analytics

iii. Antivirus and antispyware software should be used

Viruses, Trojan horses, worms, ransomware, and spyware are

all examples of malicious software that is placed on your com-
puter without your consent in order to get access to your data and
machine. Viruses can corrupt your data, slow down your computer,
or even take control of it. To avoid downloading viruses and mal-
ware in the first place, only download software from reputable
websites. Antivirus software is designed to detect and remove
viruses from your computer and incoming email. Antispyware is
sometimes included with antivirus software. To safeguard your
computer against the latest harmful software, keep your software
up to date.

iv. Safeguard all of your devices

To prevent unwanted access, password-­protect your computing

equipment, whether they be routers, PCs, laptops, tablets, or cell-
phones. Encryption should be used to store data, especially sensi-
tive or confidential information. Only keep vital information on
mobile devices in case they are stolen or lost while you are away
from home. If one of your devices is hacked, thieves could gain
access to all of your data via your cloud storage service, such as
iCloud or Google Drive.

2.10.1.3 Needs for safety

The categories of data were first discussed in this module. Individually identifiable
information (PII) or sensitive personal information (SPI) is any data on a live per-
son that can be used alone or in combination with other data to identify, contact, or
find that person. Legitimate businesses have agreements (Terms and Conditions or
Terms of Service) that allow them to use the information they acquire about you to
improve their business. Companies that utilize sensors on their own equipment or
vehicles are also legitimate users of our data. Governments and communities that
have put environmental sensors on trains, buses, or traffic signals have a right to the
data they collect.

2.10.1.4 Defending the corporate world

The module then went through security best practices. Physically guarding the out-
side and inside perimeters of places where data are housed, such as data centers, is
part of security. Because of the large number of IoT devices, their position in non-­
traditional settings, and the fact that many of them cannot be upgraded, securing
them is difficult. A hat in Black Hackers takes advantage of available Wi-­Fi on a
 Role of automation, Big Data, AI, ML IBN, and cloud computing  31

regular basis. You can take a number of steps to safeguard your company’s wireless
network. Keep your firewall set on, manage your operating system and browser, and
use antivirus and antispyware to keep your devices safe.
If you are utilizing a public or unsecured Wi-­Fi hotspot, observe these safety
guidelines: Do not access or transfer any sensitive personal information over a pub-
lic wireless network.

2.10.1.5 Personal data and devices must be kept safe

We are increasing the potential for security vulnerabilities as we install more and
more smart sensors in our houses. Sensors are frequently connected to the same
network as our home or small business equipment, allowing a breach on one device
to spread to all connected devices.

2.11 Summary

In this chapter, we discussed the latest technologies in the field of intelligent net-
works. We also discussed the roles and relationships of AI, ML, Big Data, IBN, and
cloud computing to improve the performance and QoS of networks.

References

[1] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wire-
less sensor network’. Proceedings of the 5th International Conference on
Communication Systems and Network Technologies, CSNT 2015; 2015. pp.
194–200.
[2] Kumar S., Ramaswami R., Rao A.L.N. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, IEEE, 2014. pp. 350–55.
[3] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence
unified with big data analytics, internet of things and cloud computing tech-
nologies’. 2021 5th International Conference on Information Systems and
Computer Networks (ISCON), 2021; Mathura, India, IEEE, 2021. pp. 1–6.
[4] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems; Singapore: Springer; 2020.
[5] Kumar S., Trivedi M.C., Ranjan P. Evolution of software-­defined network-
ing foundations for iot and 5G mobile networks. IGI USA: IGI Publisher;
2020. pp. 1–235. Available from https://www.igi-global.com/book/
evolution-software-defined-networking-foundations/244540
32  Intelligent network design driven by Big Data analytics

[6] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[7] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[8] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
personal communications. 2022, vol. 10(3), pp. 1–14.
[9] Haidar M., Kumar S. ‘Smart healthcare system for biomedical and health care
applications using aadhaar and blockchain’. 5th International Conference on
Information Systems and Computer Networks, ISCON 2021; GLA University,
Mathura, 22–23 October 2021; 2022. pp. 1–5.
[10] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus network-­on-­chip architecture’. International Journal of Innovative
Technology and Exploring Engineering. 2019, vol. 8(6), pp. 2278–3075.
[11] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[12] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring
using machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[13] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[14] Kumar S., Ranjan P., Tripathy M.R. ‘A utility maximization approach to MAC
layer channel access and forwarding’. Progress in Electromagnetics Research
Symposium; PIERS 2015 Prague, Czech Republic, 2015. pp. 2363–67.
[15] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
Proceedings of the International Conference on Computational Intelligence
and Communication Networks; Jabalpur, India, IEEE, 2016. pp. 79–84.
[16] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and iot for smart
road traffic management system’. 2020 IEEE India Council International
Subsections Conference (INDISCON), 2020; Visakhapatnam, India, IEEE,
2020. pp. 289–96.
[17] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. 2020 10th International
Conference on Cloud Computing, Data Science & Engineering (Confluence),
2020; Noida, India, IEEE, 2020. pp. 63–76.
 Role of automation, Big Data, AI, ML IBN, and cloud computing  33

[18] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation of
fault tolerance technique for internet of things (iot)’. 2020 12th International
Conference on Computational Intelligence and Communication Networks
(CICN), 2020; Bhimtal, India, IEEE, 2020. pp. 154–59.
[19] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. 2020 10th International
Conference on Cloud Computing, Data Science & Engineering (Confluence),
2020; Noida, India, IEEE, 2020. pp. 63–76.
[20] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 2019 4th International Conference on Information Systems
and Computer Networks (ISCON), 2019; Mathura, India, IEEE, 2019. pp.
250–55.
[21] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. IEEE; Mathura, India, 2021. pp. 1–6.
This page intentionally left blank
 Chapter 3
An intelligent verification management
approach for efficient VLSI computing system
Konasagar Achyut 1, Swati K Kulkarni 2, Akshata A Raut 3,
Siba Kumar Panda 4, and Lakshmi Nair 5

Any masterpiece is conjoined with all works of engineering, which includes the field
of computer science or an electrical and electronics or mixture of both computer
and electronics. Today, this gives the industry to understand research, evolve and
develop into newer technology unfolding many scriptures behind the engineering
works. In the similar manner, this chapter unfolds the prominent works involved in
the verification of the designs involved in VLSI domain. Considering machine learn-
ing (ML), neural networks and artificial intelligence (AI) concepts and applying
these to a wide range of verification approaches are quite interesting. The specific
kinds of Register Transfer Level (RTL) design require rigorous verification which
is targeted over any type of Field Programmable Gate Array (FPGA) or application-­
specific integrated circuits (ASICs). The verification process should be closed with
testing all possible scenarios that too with intelligent verification methods. This
chapter in the following pages brings the unique way of verification procedure
involved in the RTL development methodologies using hardware description lan-
guages. With the help of system Verilog language, the developed reusable testbench
is used for verification. The injected inputs to the testbench are randomized with
constraints, such that the design should produce accurate output. To unify the veri-
fication language, there is a dedicated methodology commonly known as Universal
Verification Methodology (UVM); by this, the chapter is extended to experience the
readers also through the coverage-­based formal verification. For continuous func-
tional verification, an intelligent regression model is also developed with the help
of ML and scripting. With this repeated injection of various test cases is possible in

1
J.B. Institute of Engineering & Technology, Hyderabad, India
2
Department of Applied Electronics, Gulburga University, Kalaburgi, Karnataka, India
3
Department of Electronics & Telecommunication Engineering, Fr. C. Rodrigues Institute of
Technology, Navi, Mumbai, India
4
Mobiveil Technologies India Pvt. Ltd, Chennai, India
5
Amrita School of Engineering, India
36  Intelligent network design driven by Big Data analytics

order to verify the functionality. Thus, with the adoption of the presented verification
environment and distinctive approach, one can affirm that the design is ready to be
deployed over the targeted semiconductor chips. As the verification is an unignor-
able procedure, this can be used to classify the algorithms developed in ML for data
clustering, data encoding and its accurate analysis. More importantly, this chapter
allows us to understand an intelligent verification model for testing the design with
regression run with the corresponding set-­up and the pass/failure analysis steps. This
structure may result in a significant reduction of the simulation time for a VLSI
verification engineer.

3.1 Introduction

With rapid growth in techniques and procedures involved in the development of

microelectronic devices for the diversified applications in the market, there have
also been a great research and design methodologies to come up with the end prod-
uct of a desired functionality through an electronic device. Within this life cycle of
the product being brought up to market end, there is a prominent stage where any
electronic device is embedded with a specific processor to perform a task based on
external inputs or feedbacks acquired from the real-­time sensors. These small and
tiny houses of transistors capable of performing such huge and complex tasks are
designed and verified by thousands of engineers in such a way that as many instruc-
tions fed to a chip in a single run are counter-­measured, analysed, get clarity over the
bugs available in the design and plan for the verification management approaches
to verify the logic design before it is brought in the world in the form of a chip. As
designing the hardware’s digital logic is a task prior of verifying the design, verifica-
tion is also as important as the design is done carefully with respect to form factors
of a chip like power, timing, area, latencies, etc. Thus to exercise these form factors
in an exhaustive way without manually giving the inputs to cover the max combina-
tions, we have several approaches in the verification stage to judge the functionality.
As specified, this chapter will take you through the taxonomy of the basic verifica-
tion methodology/plan for an RTL engineer comprising the study of SystemVerilog
and UVM models.
Once the furnishing of hardware description over RTL is completed, there are
few factors which is considered to be prominent before transiting the big code to
verification tams. These factors include timing, area and power [1]. FPGA selected
for the project has its own device factor’s consumption estimation viewed by
Electronic Design Automations (EDAs). This chapter is just focusing over the basic
procedural steps to follow over an RTL code to verify its robust, exhaustive and
regressive-­based logic performance. This brings an aim of a front-­end engineer
as to have an idea over verification structures followed in various organisations.
Verification yields two types of dividends having formal verification and dynamic
verification [2]. The major bridging gap between these two verifications is inclu-
sion of assertions and test cases during the verification environment build-­up. As to
confirm the behaviour of the design to meet the given specification, the verification
 An intelligent verification management approach  37

run is particularly known to be as the functional verification which is indeed actually

before generation of netlists and constraining the design. Once the design is synthe-
sized and layout is arranged over the fabric which brings up the long chain of routes
for fabric connectivity, in short, this gives an additional delay in logic performance;
verifying these complex designs needs a very fast method so as to communicate the
design with the test cases or reference models written in high-­level languages such
as C/C++, systemC, python, etc. In this context, the following pages will help you
understand how these high-­level languages help an interpreter actually to communi-
cate the UVM/SystemVerilog-­based environment talks to exchange the data seam-
lessly. Hopefully, we believe the readers of this chapter have a basic idea on how
libraries of OVM, UVM and SystemVerilog work with respect to the latest release
versions of these specific libraries. By this, it is to be brought in the context that this
chapter shows how an explicitly built library maintained in the cloud repositories is
also helpful in bringing up the complex test cases using high-­level languages. In this
high-­level languages being brought up in the verification environment, the following
section will allow you to understand Direct Programming Interface (DPI) used to
merge the verification language and high-­level language for verification purposes.
Also to experiment a new methodology for the generation of randomness in seed-
ing the input to a testbench environment, we are introducing an ML-­based concept
built over Python libraries so as to talk to the low-­level modules in the presence
of DPI-­based approach. Though the SystemVerilog/UVM library has the package
of generating random seeds for verifying an RTL behaviour, this basic theoreti-
cal information related to cluster of data seeding may help a verification environ-
ment to drastically identify the cluster gained at the output to be the desired values.
For this, there are few ML algorithms helping to maintain the juncture between the
Python language and the low-­level verification languages. This helps in exercising
the design regressively and exhausting the logic functionality by seeding a maxi-
mum number of combinations applying at the design under test (DUT). Apart from
the usage of Python language, this chapter takes you through a basic methodology
for bringing up the automation of verifying a design with the help of tool command
language (Tcl) scripts for the ease of understanding the actual epitome of the veri-
fication from basic methodology to the add-­ons of languages to speed up complex
RTL design verification (DV) in a single run.
This in turn helps an organisation to meet up their end tasks earlier than they
have actually scheduled for target meet. The basic and most common verification
environment followed by engineers can be seen in Figure 3.1. Here we have inter-
nal elements performing transactions at higher level of language abstraction done
by the interprocess communications modules such as mailboxes, semaphores and
events. Here we will be concentrating over how we can choose a way to build an
interface using DPI standards so as to exchange the function/tasks calling from
the program written in other language, as in Python for the generation of thou-
sands of seeds. Usually a verification environment consists of monitor, driver,
virtual interfaces, scoreboard to compare values, bus functional model (BFM),
generator and a whole environment to run all the functionalities in the class-­based
house modules.
38  Intelligent network design driven by Big Data analytics

Figure 3.1   Verification environment root block structure

The above diagram portrays the major elements a verification environment

structure has; in the context it shows the actual UVM-­based verification environ-
ment built with the class-­based modules extracted from object-­oriented program-
ming packages. Coming to its major functionality, it is done with the help of monitor
and driver the agent class is consisting of, they help to maintain the interprocess
communication to exchange the information or the seeds to be more technical in ver-
ification anomaly. Though we have a randomization facility to generate thousands
of seeds in a random sequence to be fed as inputs to the actual RTL logic under tests,
we have to make sure these are being exchanged with the outer languages with a
proper interfacing mechanism. This mainly has broader views over transaction-­level
information exchanging. For this, the industry has been giving us the vast facilities
to use the randomness method to generate and then exchange with the low-­level
languages like Verilog, VHDL, etc.

3.2 Literature study

Electronic circuits have a significant problem in terms of power usage. One strategy
to cope with this problem is to think about it early in the design phase so that it can
 An intelligent verification management approach  39

be addressed. Investigate alternative design options. Typically, a design flow begins

with a high-­level description of a whole system, requiring to produce precise model
techniques for power modelling are available, employed, allowing for the establish-
ment of a power relationship as well as other metrics. It is also necessary to take into
account techniques for characterization of power that is both efficient and effective.
This chapter’s purpose is to give a summary of the registration transfer level first.
Using power modelling and estimating approaches at the transistor level for FPGAs
and ASICs are two types of devices. The second goal is to propose a classification of
all techniques in accordance with a set of criteria [1]. For embedded systems, model-­
based system engineering is frequently used to streamline the design and develop-
ment process with verification. It enables, in particular, the modelling of system
design at a higher level of abstraction. As a result, to generate RTL code automati-
cally, a model transformation process is carried out. To accomplish this, simulation
is used for verification. The suggested meta-­model, in particular, is built on various
UVM concepts such as the scoreboard, monitor, agent and driver, among others, to
allow simultaneous modelling of test benches and system design. The proposed
meta-­applicability model is demonstrated using a memory model case study [2].
UVM is used in this chapter to provide a verification platform that can swiftly verify
the DUT. The test platform expertly employs UVM components and a flexible con-
figuration scheme, demonstrating the UVM’s many benefits. This verification plat-
form is unique as it does not rely on UVM’s reference model verification compo-
nent. This verification platform is appropriate for modules that do not have algorithm
models, such as high-­speed to low-­speed interfaces, interrupt handling modules, and
so on. The interrupt handling module realizes the full and overflow of ring first input
first output (FIFO), and the bus meets the protocol criteria, according to the results
of the verification. Moreover, the coverage rate satisfies the verification require-
ments. Chip verification is becoming increasingly crucial in the chip research and
development process, as it is a critical component of assuring the smooth tapeout of
semiconductors [3]. Because of the high demand for system-­level designs, the elec-
tronics semiconductor sector has seen remarkable expansion throughout the years
[4]. High-­performance computers, controls, telecommunications, image and video
processing, consumer electronics and other applications rely heavily on system-­
level designs. RTL represents a synchronous digital circuit in terms of stream of data
between hardware registers and the logical operative units performed on these sig-
nals in digital integrated circuit (IC) design. Each signal is presented to be a high-­
dimensional point in the toggle-­pattern matrix, which is built from value change
dumps (VCD) from a training set [5]. Few amounts of signals are automatically
selected by clustering signals with comparable switching actions, and then the
design-­specific but workload-­independent activity-­based power model is built using
regression on cycle-­accurate power traces acquired from standard CAD tools. An
FPGA is a logic device that may be modified by a client after manufacturing to con-
duct anything from simple logic gate functionalities to complicated systems on-­chip
or even AI systems. Scientific papers on Reference [6] they have found over 70 000
documents in the two major scientific databases to date. ScientoPy is utilised to do a
scientometric analysis of the top FPGA applications, encompassing papers relevant
40  Intelligent network design driven by Big Data analytics

to FPGAs from 1992 to 2018. Few among the top 150 applications were categorized
as follows: digital control, communication interfaces, networking, computer secu-
rity, cryptographic approaches, ML, digital signal processing, image and video pro-
cessing, big data, computer algorithms and other applications. DV technologies are
impacted by advancements in electronic system design. The Internet-­of-­Things and
Cyber-­Physical Systems paradigms presume devices that were immersed in physical
surroundings, have limited resources, and must provide security, privacy, dependa-
bility, performance and low-­power features [7]. In addition, a preliminary method to
multidimensional verification using ML techniques was assessed. The constrained-­
random stimulus has become widespread as a technique of stimulating a design’s
functioning and ensuring it completely satisfies expectations as ICs have become
increasingly complicated. In theory, random stimuli permit all conceivable combi-
nations to be exercised given enough time, but in fact, a completely random tech-
nique will struggle to train all possible combinations in a timely manner with very
complicated designs. As a result, steering the DV environment to generate difficult-­
to-­hit combinations is frequently required. The resulting constrained-­random tech-
nique is effective, but it frequently requires substantial human guidance to effec-
tively exercise the design in the DV context. This method produces better-­than-­random
results in a highly automated manner, allowing DV to meet its complete design
coverage goals in a shorter time frame and with minimum resources [8]. In Reference
[9], the goal of metric-­driven verification is discussed as well as its significance in
the current verification technique. It is proposed that the regression testing technique
be included in the verification environment itself, due to known restrictions. The
data gathered from the verification metrics can be utilized to fine-­tune the simula-
tion’s testing operations. This method enables dynamic manipulation of the regres-
sion testing framework and may result in significant simulation time savings. The
test segments were introduced in the presented solution. They can be launched at
any moment in the simulation that is connected to the internal state of the DUT
(checkpoint). Many new protocols are gaining traction and are now commonly uti-
lized in the business. The AMBA Advanced Extensible Interface Protocol, devel-
oped by ARM processors, includes features that were not available in the preceding
Advanced High-­Performance Protocol. The practical results of several of the proto-
col’s properties are shown in this document. It should be mentioned that verifying
such protocols is a time-­consuming and arduous operation. A verification IP already
has the essential testbench-­generating processes and may be readily combined with
other tools [10]. Verifying a design basically needs well-­prepared strategies allow-
ing many vendors to rely over a strict verification plan for the whole system [11].
Many of the negative consequences of the process, voltage and temperature changes
can be mitigated by bundled-­data designs. They described an open-­source CAD
pipeline for synthesizing block design (BD) from RTL specifications using commer-
cial tools and Perl and Tcl scripts. Edge is a flow that was created with the help of an
open-­source CPU called Plasma and Qualcomm Technologies, Inc. industrial design
[12]. There has been a lot of research into reconfigurable technology. The applica-
tion of FPGA as a developing reconfigurable platform for a controller solution was
the subject of this review. It is an embedded system that allows for a great deal of
 An intelligent verification management approach  41

freedom in the development of sophisticated ICs. The review’s three goals are as
follows: To begin, it looks into the contributions of FPGA-­based controllers rather
than ASIC-­based controllers. Second, to demonstrate the influence and success cri-
teria from past studies to improve controllers in terms of performance (response
time, complexity, flexibility and cost design). Third, to determine the optimum way
for combining these criteria to significantly increase the controller performance. To
improve factor design, the majority of studies rely on four important criteria: effi-
ciency of tuning and optimization methods, robust implementation technology to
reduce cost with more flexible design, type of technology to reduce complexity and
minimize design area and the ability of the controller to be used with different order
systems which could greatly improve controller responses [13]. Today’s technology
has advanced to the point that a system may be put on a single chip, a concept known
as system on chip (SoC). It entails microcontrollers and a variety of peripheral
devices, each of which has its intellectual property (IP) called IP cores. Various pro-
tocols such as RS232, RS422 and UART are used to create serial communication
between these IP cores. They execute point-­to-­point communication, which neces-
sitates extensive wiring and multiplexing of all bus connections to transmit data to
IP Cores. The benefit of I2C protocol is that it has a low wiring data transfer rate,
which may be increased by using ultrafast mode. The ultrafast data transfer mode is
a one-­way data transfer mode. They use the system Verilog and UVM in the tool
SimVision to verify the design of an I2C protocol between a master and a slave in
this study [14]. The design of a 3-­wire SPI protocol chip for ASICs and field-­
programmable gate arrays is presented in this study (FPGA). It is the first study to
implement the SPI protocol using VLSI and FPGA technologies for testing and veri-
fication. The functions of the SPI protocol have been successfully tested using an
oscilloscope. With only four pads, this study develops superior VLSI architecture
(including system clock in ASIC design) with cheap cost and minimal complexity
[15]. Hardware and software are designed and verified separately in a typical verifi-
cation environment. The frequency of processors is no longer growing due to
Moore’s law’s extinction. Moore’s law is no longer valid, and time-­to-­market con-
cerns are driving a new wave of technology that tightly integrates hardware and
software. This is to improve computing performance. Furthermore, data in domains
such as Data Analytics must be processed in real-­time and with minimal latency. For
applications that demand real-­time and low-­latency data processing, Solarflare’s
Application On-­load Engine is a platform that combines an FPGA processing engine
and low-­latency server. Co-­verification is not supported by the UVM, which instead
relies on the DPI for co-­simulation. Software emulation is done concurrently with
hardware verification in co-­verification when software is emulated separately on an
Instruction Set Simulator [16].
Due to the enormous number of IPs in the system that needs to communicate,
Network on Chip (NoC) has evolved as an interconnection option for modern digital
systems, particularly SoC. Various systems and routers have been implemented,
necessitating the creation of a reusable verification environment for testing single
routers as well as entire networks. They offered a reusable testing environment for
NoC platforms based on the UVM, which tests and certifies both routers and
42  Intelligent network design driven by Big Data analytics

networks in a way that can be easily modified to match different routers and net-
works. Performance factors such as injection rate, throughput and latency were also
evaluated by the environment [17]. Despite decades of effort, DV is still a costly and
time-­consuming part of the electronic system development process. With the intro-
duction of SoC architectures, verification has evolved into a multi-­platform, multi-­
tool, multi-­technology process that spans the whole life cycle. This paper presented
an instructional overview of current verification practice in the context of modern
SoC DV, current industry trends and significant research problems [18]. In contrast
to other Hardware Verification Languages (HVLs) such as Specman E, interactive
debug is not a native component of SystemVerilog simulations. We propose an
interactive debug library for UVM written in the SystemVerilog Direct Programming
Interface in this work (SV-­DPI). At its most basic level, this allows for high-­level
interactive debugging during simulation execution. This library offers the following
features: 1) writing or reading a register using the UVM register abstraction layer; 2)
generating, randomising, initialising and starting a UVM sequence on any sequencer;
and 3) calling a user-­specified SV function or task using the interactive debug
prompt. They were also shown a debugging and regression approach that outlined
the best practices for speeding up debugging and reducing regression run time.
According to preliminary findings, employing an interactive debug library lowered
debug turnaround time and regression run time dramatically. The UVM debug
library is an open-­source project that can be found on GitHub [19]. UVM stands for
Universal Verification Methodology, which is a standardized approach to testing IC
designs to achieve Coverage-­Driven Verification (CDV). To signal progress in the
DV, it combines automatic test generation, self-­checking testbenches and coverage
measurements. The CDV follows a different pattern than typical directed-­testing
methods. With the CDV, a testbench developer begins with a structured plan by
defining the verification goals. Coverage monitors, which have been introduced to
the simulation environment, are used to track progress. The non-­exercised function-
ality can be detected this way. Furthermore, the additional scoreboards reveal unfa-
vourable DUT behaviour. Three recent ASIC and FPGA projects that have
successfully incorporated the new workflow have been developed: the CLICpix2
65 nm CMOS hybrid pixel readout ASIC design; C3PD 180 nm HV-­CMOS active
sensor ASIC design and the CLICpix chip’s FPGA-­based DAQ system [20]. The
methodology aids us in improving the SoC’s performance and lowering its costs.
The most difficult task in the entire SoC is verification. Chips with multi-­million
gate designs have a higher level of complexity. In this study, the complexity issue in
SoC was handled using an advanced technique called advanced verification method-
ology. The verification task’s overall design effort is estimated to be almost 70%.
The verification methodology utilized here is primarily aimed at maximising reus-
ability for various design IP configurations in the SoC. The SoC’s time to market is
shortened thanks to its advanced reusable test bench development. The development
of a test bench for advanced verification of the SoC [21]. The SystemVerilog UVM
aims to boost verification efficiency by allowing teams to exchange tests and test
benches across projects and divisions. One of the most crucial steps in the ASIC/
VLSI design process is verification which takes up a lot of time and effort in the
 An intelligent verification management approach  43

design flow cycle to ensure that the design is bug-­free. As a result, a powerful and
reusable verification approach is in high demand. UVM is based on various method-
ologies such as verification methodology manual (VMM), open verification meth-
odology (OVM) and e reuse methodology (eRM). It can be used to verify designs
written in a variety of languages, including Verilog, VHDL and System Verilog.
UVM allows for a reusable verification environment, which saves time during the
verification cycle [22]. Software models are used as a golden reference model for the
algorithm after it has been finalize. They described a unified and modular modelling
framework for image signal processing algorithms that can be used for a variety of
applications, including ISP algorithm development, reference for hardware imple-
mentation, reference for firmware implementation and bit-­true certification. The
functional verification framework of image signal processors utilising software ref-
erence models is based on the UVM. IP-­XACT-­based solutions for the automatic
production of functional verification environment files and model map files are also
well described [23]. Technical publications frequently make subjective or unfounded
assertions regarding today’s functional verification process, such as verification
consumes 70% of a project’s overall work. Despite this, there are few trustworthy
industry studies that quantify the functional verification process in terms of verifica-
tion technology uptake, effort and efficacy. To overcome this knowledge gap, a
recent global, double-­blind functional verification study covering all electronic
industry market categories was conducted. This is the largest independent functional
verification research ever undertaken to our knowledge [24]. FPGA technology has
progressed significantly. Some advancement may be in terms of previously avail-
able resources, while others may be in terms of overcoming traditional limitations or
improving efficiency. The transition in focus to soft core/embedded processors has
resulted in the transformation of FPGA from just hardware to a powerful System-­
on-­Chip. IP cores and design tools have also progressed, with software developers
providing improved support. The emergence of new FPGA capabilities on various
industrial applications is examined in three primary areas: digital real-­time applica-
tions, sophisticated control approaches and electronic instrumentation, with a focus
on mechatronics, robotics and power electronics [25]. Because of its great efficiency
and low power consumption, FPGA has become popular in new big data architec-
tures and systems, allowing researchers to install enormous accelerators on a single
chip. In this work, they propose the software defined accelerator (SODA) technol-
ogy, which is a software-­defined FPGA-­based accelerator that aids in the re-­
construction and re-­organization of acceleration engines based on the needs of
various computation applications. This SODA is made up of several layers. Large
and complicated applications are decomposed into coarse-­grained single-­purpose
RTL code libraries that perform specialised tasks in out-­of-­order hardware [26].
According to Moore’s law, hundreds of gates are being added to the SoC architec-
ture as technology shrinks. Meanwhile, getting a product to market is growing more
complicated. Due to the rising complexity of SoC, verification has become more
difficult. Because current methods are unable to handle the efficiency of growing
design sizes, there is a need for innovation [27]. With the advent of multi-­core com-
puters, parallelizing performance-­critical applications of many types has become a
44  Intelligent network design driven by Big Data analytics

difficulty. Parallel computation aids in the division of big and difficult jobs into
smaller units, each of which is handled by a different processor. This aids in improv-
ing performance and completing difficult jobs. A customizable parallel device
known as a FPGA remains a minor area of study in parallel computing. The FPGA
was discussed in this research study as a technique for improving parallel applica-
tions by functioning as a co-­processor rather than conventional CPUs. The parallel
architecture of an FPGA allows complex operations to be moved from the CPU and
into specially designed logic inside the FPGA, resulting in excellent performance at
a low cost. FPGAs are useful as development tools and are more cost-­effective for
computing applications. As a result, for a surging number of higher volume applica-
tions, FPGAs have shown to be a better option to traditional CPUs [28]. The overall
development process becomes increasingly critical and demanding as the design
complexity of SoC verification grows. UVM is a standard solution to SoC design
complexity, even though there are still unsolved issues. The future of SoC verifica-
tion technique is discussed by specialists from business and academia in this publi-
cation [29]. Ultra-­High Frequency (UHF) radio frequency identification (RFID) is a
fast-­growing technology that uses RF signals to automatically identify things. RFID
is currently being used for a variety of applications, including vehicle tracking and
security, as well as asset and equipment tracking. Due to the module-­reuse strategy
and low-­power techniques used in the digital baseband, the RFID tag’s power con-
sumption is minimized. For verification, an intelligent and adaptable testbench
architecture based on UVM is developed. To improve efficiency and provide auton-
omous stimuli, complete algorithms for pathfinding (CAP) based on state transition
graph are employed [30]. SoC verification is one of the popular trends in VLSI right
now. Verification takes up most of the time. As a result, it was necessary to create a
usable and efficient verification environment. UVM is a SoC functional verification
methodology. UVM explains how to verify an IP and set up a productive verification
environment. To distinguish between UVM-­based and traditional verification,
research was conducted [31]. The major reason for functional verification on hard-
ware is to discover issues in the designers’ design specifications and to evaluate the
design functioning. The design is then tweaked as needed to achieve the intended
functionality of the DUT. There are a variety of verification strategies that make the
process more user-­friendly. SystemVerilog is a huge and complex programming
language. UVM is a powerful and versatile class library that has evolved over time
[32]. UVM includes powerful SystemVerilog capabilities. It includes a reference
verification mechanism in addition to a robust base class library. It was investigated
whether the addition of certain SystemVerilog features aided in the evolution of
UVM. The combination of UVM and SystemVerilog has been found to provide
users with a comprehensive toolbox that can be used to solve a variety of challenges
in the functional verification domain of hardware designs [33]. The Ninth Haifa
Verification Conference HVC 2013 [34] organized by IBM Research was to advance
the state-­of-­the-­art and profession in verification and testing. A forum of industry
experts, academic practitioners and researchers gathered here to share their work,
exchange ideas and explore future testing and verification prospects for hardware,
software and complex hybrid systems. One of the current difficulties in the
 An intelligent verification management approach  45

pre-­silicon validation area is the lack of a single framework for all pre-­silicon valida-
tion efforts, including security validation. As a result, multiple validation tool teams
are having trouble communicating. As each team works to improve a specific tool
for the project they serve, the project’s geography shifts to keep up with the product
development life cycle, tool updates and addressing specific tool adjustments, all of
which must be communicated among several teams [35]. UVM is a verification
approach for the IP and SoC. This is based on the SystemVerilog HVL and allows
verification components to be reused. In complex systems, SoC verification is diffi-
cult because it involves a lot of effort to validate numerous IP interface and on-­chip
inter-­IP interaction. The reusability of IP verification environment is a desired but
difficult approach [36]. The degree of complexity for component and system design
for functional verification increases as the design and sophistication of Multiprocessor
SoC designs grow to meet minimal power, speed, performance and functionalities.
The objective of verification is to guarantee that the designed system is operating in
accordance with the specified requirements. In this study, a flexible verification
environment is introduced that adapts to the associated NoC framework [37]. Many
obstacles were encountered in prior verification approaches, including reusability,
maintenance, bug identification, and so on. The introduction of SystemVerilog
UVM overcomes these obstacles. At the SoC level, where each IP is a black box and
regarded as a golden block, IP level verification is an important factor. In this
research, a set of reconfigurable image signal processing IPs is combined to meet the
needs of a variety of advanced video processing SoCs [38]. Verification activities
have begun to dominate the design effort as systems get more complicated. More
efficient verification languages and tools have been devised to cope with rising com-
plexity and enhance productivity at the ESL (Electronic System Level). UVM is a
methodology for RTL and ESL verification that includes a complementary library.
SVM with extensive TLM support based on System C was introduced in this chap-
ter, and it was compared to UVM-­related features [39]. The software drivers are
produced only once the hardware is available. This reliance can result in a prolonged
design cycle. It is not always easy to specify driver development and hardware/
software interface models. A technique to formalize hardware/software interface
requirements where concurrency is required is presented in this work, and this
approach is proven using a practical scenario and elaboration of use [40]. This chap-
ter provides an overview of programmable hardware trends in the semiconductor
industry. FPGAs have become synonymous with development and reprogrammable
computing. This chapter focuses on many new industry developments as well as
FPGA, with the goal of being informative. The reasons for the emergence of differ-
ent technologies, their benefits and drawbacks, and why FPGA is the dominating
choice when it comes to programmable logic are all explored here. There is also
debate about whether these technologies will continue to exist [41]. Building a test-
bench to verify a hardware design is referred to as a software challenge. The basic
elements of a design, namely registers, linkages between them, and the computation
required to adjust their values, are all included in the RTL. The procedure for creat-
ing an RTL design is similar to programming. In the software world, testbenches
survive. Data structures and algorithms make up the testbench. Even though most of
46  Intelligent network design driven by Big Data analytics

their creation and operation is software, they are hardware-­aware because it is their
role to regulate, adapt to and evaluate the hardware [42]. There is a desire to com-
pare FPGAs and ASICs in terms of space, power consumption and performance.
The scientific evidence characterizing the gap between FPGAs and ASICs are pre-
sented in this work. When compared to an identical ASIC, it was discovered that the
FPGA takes up more space, performs slower, and consumes more power. It was
eventually verified that using hard multipliers and dedicated memory blocks can
save a significant amount of space and power while only having a modest influence
on the latency between FPGAs and ASICs [43]. The industry’s need for complicated
and efficient high-­performance controllers prompted rapid development of VLSI
technology and EDA methodologies. EDA tools are used to generate, simulate and
test a design without committing to hardware, allowing complicated systems and
concepts to be quickly evaluated. Because of the increasing difficulty of control
algorithms and chip density, efficient design methodologies are required [44]. DV
has surged in popularity in the electronics industry. It began with the shift in verifi-
cation requirements from ICs to SoCs and bigger systems, which was amplified by
the rise in embedded processor utilization. Then it went on to expand the quantity
and range of verification approaches and languages available. Finally, books con-
taining instructions and advice on verification methodologies, languages, and tools,
as well as how to combine all of them into a unified verification approach and pro-
cess, have become more widely available [45]. As ASIC designs get more compli-
cated, the complexity of the ecosystems in which they were implemented rises
exponentially. There has been no parallel robust procedure in place for verification
environments like an ASIC design’s HDL may be subdivided into intelligible but
small components whose behaviour can be grasped in a reasonable amount of time.
Any verification environment established or generated for these design sub-­blocks,
whether written in HDL or any of the different verification or scripting languages
now accessible, remains extremely complex. The goal of the team was to combine
C++, Tcl and Perl into a unified, highly intelligent and usable ASIC verification
environment [46–67].

3.3 Verification management approach: Case Study 1

The forthcoming [64] industrial revolution has the greatest man-­made theories yet
to be proven in actual practical scenarios, though the intention of the technology
is to reduce the manpower and make use of the so-­called computers at a greatest
extent. When we make a note of the technology called “computer,” we tend to aim
towards the development of both hardware as well as software productions and yet
here we are relating to one such field of computers where the core of its opera-
tions is embedded over a single fabric with the lightest and more and more thinner
technology nodes developing by huge multinational companies. This in turn again
gives an end user to reduce the man power and yet depend over the software so as to
develop/demonstrate/simulate/calibrate an actual hardware which has to be brought
up from the idea created to the end market for the consumers. ML is one such end
 An intelligent verification management approach  47

where computers are used at its full potential which is again to study the computer
algorithms with the help of the usage of huge amount of data nodes. Data nodes is
a term referring to bunch of information pointing to one subject and these subjects
may differ from point to point. To access these nodes or to identify the information
in a very minute period of timescale, we need faster developed algorithms, as fast
as it could also be utilised to judge a qubit (quantum bit) in the quantum computing.
Having this phenomenal facility of ML, this chapter possesses the strategy to imple-
ment the randomization technique with the help of xorshift128 algorithm which is
a part of ML. Internal to this, we have to keenly observe the network topology of
the data redundancy and node accessibility, and this stage is well known to be the
neural networks.

3.3.1 The pseudo random number generator in a verification

environment
Basically machine learning is mainly used to identify the sequence of data being
generated and let the perfect output with respect to the clusters collected and exam-
ined; this has a numerous stages of repeated processes in a neural network termi-
nology. Deep learning is also one such ML task where huge amount of information
is collected and passed throughout several algorithms to extract the exact output.
Pseudo random number generator (PRNG) way of addressing the seeds towards the
test/verification environment could yield us as most of the 95% of accuracy are from
the processed end results.
Coming toward the xorshift128 algorithm, it is one such specific task or operation
implemented pretty straightforward. Basically it has four internal variables suppose w,
x, y and z representing the seed simultaneously. Instead of just hardcoding, we have
special packages for the generation of these seeds. Each time the generator function has
been called, this will shift the variables from y to x, z to y and w to z. Here in this context
the variables are representing the actual binary values to be seeded or randomized at
instance. The new value generated at w is now again executed by bit manipulations like
left/right shits and XORing to the past values of w, x, y and z. From this we can observe
that to generate one output suppose o1, we need just two of the variables. No doubt that
the other variables are also utilised in the further stages of the randomization, but at
this instance, it is to be noted that for just one output variable we need lesser amount of
input variables. As this PRNG algorithm is used in this context just to generate random
amount of bit patterns, it is also important to note that this algorithm is cryptographically
not highly secured. Now to implement this algorithm digitally is possible by introducing
Linear Feedback Shift register (LFSR), but the delay has to be maintained as equal to the
class-­based modules as the transactions are being taken up to the higher level languages
and it is undesirable to miss any sort of combination generated by LFSR modules in
digital significance for verifying a specific design. As it has been mentioned about the
timing, this parameter is not that closely observed during functional verification stage
also it has to be made sure that high-­level language is independent of time-­consuming
constructs and task specifications.
48  Intelligent network design driven by Big Data analytics

Figure 3.2  DPI linkage with actual scenario (PRNG and floating point
conversion module)

In a statistical sense, xorshift algorithm generates a random number only when it

has the status of the present state but it is purely unknown to a user. But the sequence
it has generated by an ordinary pseudorandom generator often allows one to determine
the internal state of the algorithm/generator. This in turn observing the fair long enough
generated sequence allows a user to predict the future sequences in most of the cases. In
this chapter, we have taken xorshift128 algorithm which consists of 2128–1 period and
passes all the possible combinations on a huge stream of bits for a burst transfers. In a
single task, this algorithm allows a user to generate a sequence of multiple random num-
bers. As in Figure 3.2 after generating a random number, we can see that this is followed
by the single precision floating point converter which helps the digital logic to determine
the signed and unsigned integers to follow the complex algorithms written in RTL.
Multilayer perceptions are the most prominent units in ML algorithms which are uni-
fyingly known as neurons which combined with thousands of such units and technically
 An intelligent verification management approach  49

Figure 3.3   Internal structure of xorshift network

known to be a neural network. From Figure 3.3 as we are able to vision, the internal struc-
ture of the xorshift algorithm having four input seeds with the range ‘n’ is shifted multiple
times with the other seed bits and is exported to form a new set of output seeds for the
random bit generation.

3.3.2 Implementation of PRNG in higher abstraction language and

usage of DPI
It makes sense now that the xorshift algorithms used to drive the PRNG algorithm
use the sequence of internal states/patterns of bit streams to generate the future
sequence. Now as this chapter is portraying the eccentric way of utilising these
higher level languages in the RTL verification environment, it is also important to
know how DPIs actually work and finally what are the essential elements to initial-
ize our PRNG algorithm into the Python-­based script so as to seed the input for the
verification environment of a specific RTL logic. The reason to choose the bit length
for xorshift as 128 bits is because as per Figure 3.3 we have four outputs which are
32-­bit wide each.
50  Intelligent network design driven by Big Data analytics

Now as verification of a digitally described logics with the help of standard

RTL languages seldom requires a need to communicate with the higher abstraction
languages such as C, C++, systemC, Python, etc., this task is not done at ease just be
instantiating the mutual variables and classes. During this special operation, there
are several layers of abstraction where a system has to undergo several handshaking
sort of mechanism which allows Python to be communicating with SystemVerilog.
The first and the foremost layer which is way too important to acquire the languages
written to communicate is the end user layer; this layer is responsible for initiating
the transaction between SystemVerilog and other code, for instance, Python. This
can be done by using specific function “call_client()” using system address and
serial of port of the server. Followed by we have the connection layer which acts
as the actual media between SystemVerilog and Python language. This can also be
well known as the bridging layer between two different languages and this task is
performed in the client-­server mode. In this layer, there are several handshaking
mechanisms to be followed where the end user and the actual server are having the
mutual TCP/IP address to identify the socket for communication. The third one is
the client layer which acts as the agent between the end user layer and the server
layer. In this layer, a connection/transaction is initiated within this. With this, the
end user (client) and the server are connected using the DPI-­C. Basically to call
the functions from C library into SystemVerilog, we have DPI-­C. This makes sure
that the data exchanged at both the ends are interpreted up to mark and are exactly
the same. Thus to import the function, we have a straightforward keyword “import
DPI-­C” which can be written in SystemVerilog codes to extend the functions usage
like test_bit, transform_bit, etc. Now to import the DPI features into C program as
this forms a basic layer in between end user (client) and server knowingly DPI-­
C, we have a package in C language too where the keywords are prescribed as
“#include svdpi.h.” These things are well maintained in the client layer, apart from
this connection errors and data incoherency are also well prompted in this layer as
in Figure 3.4. Finally, we have the server layer which is the actual oasis for the new

Figure 3.4   Block view of DPI and different languages on either side
 An intelligent verification management approach  51

language which needs to talk to our SystemVerilog code. Basically, this layer is also
responsible for providing a clean response to a client from time to time. Here, the
message is well decoded and received from the end user and generates the informa-
tion to be reverted back to the client (end user). With this one successful transaction
with client, the server is now responsible to close the connection indicating that
the information has been successfully exchanged between the connection layer and
vice versa. Usage of DPI directly in Python can be done by calling “pythonbind11”
package bit; this has a disadvantage of not cohesively performing actions with the
communication layers in between the client and server, making it hard to debug the
code and its functionality.
Bridging of crucial high-­level abstraction and low-­level described languages
needs a strong and precise interpreter in terms of computer science theory, as the
mutual compilation feedbacks are necessary so as to spawn the exact output from
the interfacing components used.

3.4 Verification management approach: Case Study 2

Tcl is a dynamic programming language that is powerful and simple to learn. It

is suited for various applications, including online and desktop applications, net-
working, administration, testing, and so on. Tcl is an open-­source and business-­
friendly language that is truly cross-­platform, readily deployed and highly flex-
ible. As one of the most complex SoC, FPGA and ASIC development flow steps,
verification requires the development of novel methodologies. The following is
an overview of Tcl compared to SystemVerilog and VHDL. Tcl is not the same
as HDL or HVL! It is a general-­purpose scripting language. Tcl is used for vari-
ous practical purposes like scripting, DSL (domain-­specific language) generation,
testing and verification. Tcl is widely used throughout the industry as an embed-
ded EDA and CAD language – it is utilized for automation, testing, configuring,
“glue” languages and nearly every programming duty necessary in verification.
It is preferable to work closely with synthesis/compile tools simulators, which
provides significant benefits. But what do we know about it? Tcl expertise is not
as widespread as C knowledge.
What are the critical differences between general-­purpose Tcl, SystemVerilog
and VHDL for verification? Tcl, unlike HDLs, operates outside of the simulator via
the interpreter included with the EDA tool. It manages the simulator by interpreting
scripts based on predefined commands – without the need for compilation. Console
and GUI mode can control the simulator at several levels. But, most importantly, it
is feasible to install a simulator inside and maintain a DUT or a DUT with a testing
environment (UVM or another) in the same way.
Built-­in vendor support:
52  Intelligent network design driven by Big Data analytics

•• Mentor Modelsim/Questa – Tcl-­based – over 350 well-­documented commands

in the “Command Reference Manual” and complete internals control is possible
most of the time.
•• Synopsys VCS – embedded Tcl – complete UCLI and DVE control
•• Cadence Incisive – just one quote from the website: “For Electronic Design
Automation (EDA) and Computer-­Aided Design (CAD) applications, Tcl has
become the de facto standard embedded command language.”

UVM support, command extension to C/C++code and configurations, whereas

Tcl is a powerful and versatile tool ideal for various situations and standard verifi-
cation methods. The inbuilt Tcl interpreter in the simulator provides full language
support for verifying clear DUT, expanding current test benches and working with
varied settings.
The benefits of using Tcl for verification

1. It is possible to “edit, execute, stop and restart” Tcl scripts without recompiling
or exiting the simulator.
2. Because of the nature of Tcl, it is far more expressive and prolific than
SystemVerilog.
3. DSL support from scratch.
4. Flexible creation of any data structures
5. Automation and enhancements for simple static analysis
6. The most straightforward method for interacting with external simulator EDA/
CAD tools.
7. An easy way to communicate with something written in another language.
8. There is no need for anything additional because Tcl is already included in EDA
tools.

Due to language limitations, once a testbench is created in SystemVerilog, the

user has little influence over the test scenario during simulation. Users are currently
limited to using a different seed or poking/forcing signals within the simulation
environment during simulation run time.
Consider the following situation: A user has a lengthy simulation that takes
hours to complete. A testbench checker flags an error halfway through the simula-
tion. If the user has to alter the stimulus to troubleshoot the problem, they must
update the SV code, recompile the testbench and wait for a long time to identify the
error again. The interactive debug library allows users to experiment with different
debug circumstances without restarting the simulation. They can query the device
status via register reads, try new device configurations via register writes, adjust
the stimulus generated by the testbench by performing a new UVM sequence with
a higher priority or invoke any testbench functions to assist in the identification of
the problem.
They can test dozens of scenarios and find the bug in minutes rather than hours.
The significant reduction in debug turnaround time increases productivity by isolat-
ing the bug rather than waiting for the simulation to compile/run.
 An intelligent verification management approach  53

‍ ‍

The debug prompt helps experiment with different test scenarios in real time. If
the same set of debug instructions is used repeatedly, loading a command file saves
time on typing. The command and descriptions are given in Table 3.1.
On the other hand, the debug prompt lacks any built-­in programming fea-
tures and can only execute one command after another consecutively. To put it
another way, the debug prompt is not a Turing complete language in and of itself.
Fortunately, the simulator already includes a full-­featured programming environ-
ment in the shape of the Tcl prompt.
The UVM debug library produces a Tcl wrapper function that calls the debug
prompt, passes in and executes one debug command, and returns to the Tcl prompt
with the debug command’s return value.
Please remember that the Tcl script integration is currently in its early stages.
Everything is executed in a single thread by the Tcl prompt.
It would be interesting to observe how well the UVM debug library performs
when several threads call numerous debug instructions simultaneously.

Table 3.1   Command basis approaches

Command Description
Help [command] Displays the help message. With no argument, it lists all
the available commands; otherwise it displays the help
message of the command specified
Continue Exits the debug prompt and continue the simulation
pause Pauses the simulation and switch to the simulator’s Tcl prompt
Run < runtime > Runs the simulator for the specified time and go back to the
debug prompt
History [list] |clear |save< Lists all the previous command inputted by the user; clear the
file > history; save the history to a command file
Repeat # Repeats the specified line in the command history
Read < file > Reads a command file
Save_checkpoint [-path< Saves a check point snapshot
path > ]< snapshot
name>
54  Intelligent network design driven by Big Data analytics

3.5 Challenges and research direction

We address the new challenges and directions that big data and AI pose in education
research, policymaking and industry. Big Data and AI applications in education have
made tremendous strides in recent years. This exemplifies a new trend in cutting-­
edge educational research.

3.5.1 Challenges in intelligent systems

Research in intelligent systems faces numerous challenges, many of which relate to
computationally representing a dynamic physical world.
Inject the faults/errors: One of the most difficult challenges is determining
how to inject the faults/errors.
For external faults, the stimuli delivered by the testbench to the DUT must be
adjusted. Internal flaws require errors to be introduced into the DUT, but the design
should not be altered.
Code coverage: Coverage is the most significant parameter of the verification
process, and it is meticulously recorded during each test run in the traditional man-
ner. Coverage collecting adds a substantial amount of time to the simulation time. It
is sufficient to collect coverage for simply a subset of modules. The coverage level
for the remaining modules may be correctly predicted using Deep Neural Networks.
Grouping of Test cases: ML techniques (which offer a means of implementing
AI) can be effectively utilised to classify the causes of test failures. It is also feasible
to group failing tests based on the reason for failure, although the results on that
problem did not match the expected outcome.
Uncertainty: Physical sensors/effectors deliver limited, noisy, and erroneous
data/action. As a result, any actions taken by the system may be wrong, both due to
noise in the sensors and to restrictions in performing those actions.
Dynamic world: The physical world is constantly changing, necessitating judg-
ments made on short time scales to account for environmental changes.
Time-­consuming computation: Finding the best path to a goal necessitates a
lengthy search through a massive state space, which is computationally expensive.
The disadvantage of devoting excessive time to computing is that the world may
change in the interim, rendering the computed strategy obsolete.
Mapping: A significant amount of information is lost during the transition from
the 3D world to the 2D world. Changes in perspective, lighting, and scale; back-
ground clutter or motion; and grouping things with intra-/inter-­class variations are
all issues for computer vision.
In hardware verification, AI is applied. The specific language feature enables
verification engineers to dynamically load coverage reports from prior runs into a
test (to make the verification environment behave in accordance with coverage level
%). Which e-­language structures could adequately fulfil the necessity for giving
relevant data to ML algorithms? Typically, the tests in a verification environment
are grouped into one or more regressions and performed several times until the veri-
fication metrics are met. The current study is crucial in incorporating AI concepts
 An intelligent verification management approach  55

into the IC verification process. ML in verification can be utilised in a variety of

ways, including producing code for driving the full range of stimuli required to get a
verified circuit into all of its functional states, assessing test results, fixing coverage
gaps, and so on. However, a significant difficulty is determining which portions of
the verification environment can best address automation objectives and which veri-
fication methodologies are most suited for giving data to automatic code generators.
Some characteristics of e-­language are mentioned in this chapter as welcome bridges
between hardware verification and AI. Standard UVM is currently implemented in
SystemVerilog. However, several other languages, such as MATLAB/Simulink®,
VHDL and SystemC, are used for electronic system design. When selecting a lan-
guage that is best suited for a specific application, it is advisable to always use the
same verification approach and testbench library, as with C-­based implementations.

3.6 Conclusion

Verification environment itself is a vast terra firma where thousands of verification

plans and strategies meet the design specifications functionality and its pertained
output is well scrutinised by coverage reports and design meeting the planned veri-
fication reports. So does the industry is on the verge of utilising many plans to meet
the time to market strategy for faster consumer needs. In this chapter, a couple of
separate methods have been brought where one can effectively utilise the trending
ML algorithms with the help of high-­level programming languages like Python for
instance, to speed up the interfacing junction and to receive the data in a surplus
amount with fraction of amount of time. This chapter has taken you through just
the ideology for implementing the verification plan in a brisk through manner with
the help of DPI. UVM architecture is well reserved for its root of ahead-­plans to be
maintained over different DUTs, so is the SystemVerilog’s huge applicable concepts
and language syntaxes act as a point of source for understanding UVM and making
it conjoin with the methodologies described in the above sections making a drastic
linkage in between high-­processing algorithm with large data cluster identification
and its randomisation conceptuality.

References

[1] Nasser Y., Lorandel J., Prevotet J.-C., Helard M. ‘RTL to transistor level pow-
er modeling and estimation techniques for FPGA and ASIC: a survey’. IEEE
Transactions on Computer-­Aided Design of Integrated Circuits and Systems.
2021, vol. 40(3) 479–93.
[2] Anwar M.W., Qamar S., Azam F., Butt W.H., Rashid M. ‘Bridging the gap
between design and verification of embedded systems in model based system
engineering’. A Meta-­model for Modeling Universal Verification Methodology
(UVM) Test Benches’. ICCMS; 2020. pp. 26–8.
[3] Zhou S., Geng S., Peng X., et al. ‘The design of UVM verification platform
based on data comparison’. Proceedings of the 4th International Conference
56  Intelligent network design driven by Big Data analytics

on Electronic Information Technology and Computer Engineering (EITCE

2020); Xiamen China, 6–8 Nov; 2020. pp. 1080–5.
[4] Singh D., Chandel R. ‘Register-­Transfer-­level design for application-­specific
integrated circuits’. Nanoscale VLSI. 2020.
[5] Kim D., Zhao J., Bachrach J., Asanović K. ‘Simmani: runtime power mod-
eling for arbitrary RTL with automatic signal selection’. 52nd Annual IEEE/
ACM International Symposium on Microarchitecture (MICRO '52); New
York; 2019. pp. 1050–62.
[6] Ruiz-­Rosero J., Ramirez-­Gonzalez G., Khanna R. ‘Field programmable gate
array applications—a scientometric review’. Computation. 2019, vol. 7(4), p.
63.
[7] Lai X., Balakrishnan A., Lange T., et al. ‘Understanding multidimensional
verification: where functional meets non-­functional’. Microprocessors and
Microsystems. 2019, vol. 71,102867.
[8] Hughes W., Srinivasan S., Suvarna R., Kulkarni M. ‘Optimizing design
verification using machine learning: doing better than random’. ArXiv.
2019,arXiv:1909.13168v1.
[9] Cieplucha M. ‘Metric-­Driven verification methodology with regression man-
agement’. Journal of Electronic Testing. 2019, vol. 35(1), pp. 101–10.
[10] Bera R., Sarkar S., Singh O., Saikia H. (eds.) Advances in communication,
devices and networking. Vol. 537. Singapore: Springer; 2018. Available from
https://link.springer.com/book/10.1007/978-981-13-3450-4
[11] Mehta A.B. ‘ASIC/soc functional design verification’ in ASIC/soc functional
design verification A comprehensive guide to technologies and methodolo-
gies. Springer Nature Switzerland: Springer Publications; 2018.
[12] Zhang Y., Cheng H., Chen D., et al. ‘Challenges in building an open-­source
flow from RTL to bundled-­data design’. 24th IEEE International Symposium
on Asynchronous Circuits and Systems (ASYNC); Vienna, Austria, 13–16
May; 2018. pp. 26–7.
[13] Alkhafaji F.S.M., Hasan W.Z.W., Isa M.M., Sulaiman N. ‘Robotic controller:
ASIC versus FPGA—a review’. Journal of Computational and Theoretical
Nanoscience. 2018, vol. 15(25), pp. 1–25.
[14] Satapathy S., Bhateja V., Das S. (eds.). ‘High Level Verification of I2C
Protocol Using System Verilog and UVM’. Smart Computing and Informatics.
Smart Innovation, Systems and Technologies. 78. Singapore: Springer; 2018.
[15] Tuan M.-C., Chen S.-L., Lai Y.-K., Chen C.-C. ‘A 3-­wires SPI protocol CHIP
design with application-­specific integrated circuit (ASIC) and FPGA verifica-
tion’. Proceedings of the 3rd World Congress on Electrical Engineering and
Computer Systems and Science (EECSS'17); Rome, Italy; 2017.
[16] Birla S., Sharma S., Shukla N.K. ‘UVM-­powered hardware/software co-­
verification’. Journal of Information and Optimization Sciences. 2017, vol.
38(6), pp. 945–52.
[17] Eissa A.S., Ibrahem M.A., Elmohr M.A. ‘A reusable verification environment
for NoC platforms using UVM’. IEEE EUROCON 2017 -17th International
Conference on Smart Technologies; Ohrid, Macedonia, 6–8 Jul; 2017.
 An intelligent verification management approach  57

[18] Chen W., Ray S., Bhadra J., Abadir M., Wang L.-C. ‘Challenges and trends in
modern SOC design verification’. IEEE Design & Test. 2017, vol. 34(5), pp.
7–22.
[19] Chan H. ‘UVM interactive debug library: shortening the debug turnaround
time’. DVCON 2017; San Jose; 2017.
[20] Fiergolski A., on behalf of the CLICdp collaboration. ‘Simulation environ-
ment based on the universal verification methodology’. Topical Workshop on
Electronics for Particle Physic; Karlsruhe, Germany; 2016.
[21] Renuka G., Ushashree V., Chandrasekhar Reddy P. ‘Functional verification of
complex SOC by advanced verification methodology’. International Journal
of VLSI System Design and Communication Systems. 2016, vol. 4(12), pp.
1308–12.
[22] Pankaj S.V., Kureshi D.A.K. ‘UVM architecture for verification’. International
Journal of Electronics and Communication Engineering & Technology. 2016,
vol. 7(3), pp. 29–37.
[23] Jain A., Gupta R. ‘Unified and modular modeling and functional verification
framework of real-­time image signal processors’. VLSI Design. 2016, vol.
2016 1–14.
[24] Foster H.D. ‘Trends in functional verification: a 2014 industry study’. 52nd
ACM/EDAC/IEEE Design Automation Conference (DAC); San Francisco,
CA, IEEE, 2015. pp. 1–6.
[25] Rodríguez-­Andina J.J., Valdés-­Peña M.D., Moure M.J. ‘Advanced features
and industrial applications of FPGAs—a review’. IEEE Transactions on
Industrial Informatics. 2015, vol. 11(4), pp. 853–64.
[26] Wang C., Li X., Zhou X. ‘SODA: software defined FPGA based accelerators
for big data’. Design, Automation & Test in Europe Conference & Exhibition
(DATE); IEEE, 2015. pp. 884–87.
[27] Oddone R., Chen L. ‘Challenges and novel solutions for soc verification’.
ECS Transactions. 2014, vol. 60(1), pp. 1191–5.
[28] Opeyemi M.A., Justice Emuoyefarche E.O. ‘Field programmable gate array
(FPGA): a tool for improving parallel computations’. International Journal
of Scientific & Engineering Research. 2014, vol. 5(2), pp. 2229–5518.
[29] Drechsler R. ‘Panel: future soc verification methodology: UVM evolution or
revolution?’. Design, Automation & Test in Europe Conference & Exhibition
(DATE); IEEE, 2014. pp. 1–5.
[30] Li Q., Xie Z., Su J., Wang X. ‘UVM-­based intelligent verification method
for UHF RFID tag’. IEEE International Conference on Electron Devices and
Solid-­State Circuits; Chengdu, China, 18–20 Jun; 2014. pp. 1–2.
[31] Salah K. ‘A UVM-­based smart functional verification platform: concepts,
pros, cons, and opportunities’. 9th International Design and Test Symposium;
Algeries, Algeria, 16–18 Dec; 2014.
[32] Raghuvanshi S., Singh V. ‘Review on universal verification methodol-
ogy (UVM) concepts for functional verification’. International Journal
of Electrical, Electronics and Data Communication. 2014, vol. 2(3), pp.
101–07.
58  Intelligent network design driven by Big Data analytics

[33] Bromley J. ‘If system verilog is so good, why do we need the UVM? sharing
responsibilities between libraries and the core language’. Proceedings of the
2013 Forum on specification and Design Languages (FDL); Paris, France,
24–26 Sep; 2013.
[34] Valeria Bertacco Axel Legay ‘Hardware and software: verification and test-
ing’. 9th International Haifa Verification Conference; Haifa, Israel, Springer
Cham, 2013. pp. 5–7.
[35] Kannavara R. ‘Towards a unified framework for pre-­silicon validation’.
Proceedings of the 4th International Conference on Information, Intelligence,
Systems and Applications; Piraeus, Greece, IEEE, 2013. pp. 321–26.
[36] Zhaohui H., Pierres A., Shiqing H, et al. ‘Practical and efficient SOC veri-
fication flow by reusing IP testcase and testbench’. 2012 International SoC
Design Conference (ISOCC); Jeju, Korea (South), IEEE, 2012. pp. 175–78.
[37] Lim Z.N., Loh S.H., Lee S.W., Yap V.V., Ng M.S., Tang C.M. ‘A reconfigur-
able and scalable verification environment for noc design’. Conference on
New Media Studies (CoNMedia); Tangerang, Indonesia, IEEE, 2013. pp. 1–4.
[38] Jain A., Bonanno G., Gupta D.H., Goyal A. ‘Generic system verilog uni-
versal verification methodology based reusable verification environment for
efficient verification of image signal processing IPS/SOCS’. International
Journal of VLSI Design & Communication Systems. 2012, vol. 3(6), pp.
13–25.
[39] Oliveira M.F.S., Kuznik C., Mueller W., et al. ‘The system verifica-
tion methodology for advanced TLM verification’. Proceedings of the
Eighth IEEE/ACM/IFIP International Conference on Hardware/Software
Codesign and System Synthesis (CODES+ISSS); Tampere Finland, 7–12
Oct; 2012.
[40] Li J., Xie F., Ball T., Levin V., McGarvey C. ‘Formalizing hardware/soft-
ware interface specifications’. 26th IEEE/ACM International Conference on
Automated Software Engineering (ASE 2011); Lawrence, KS, IEEE, 2011.
pp. 143–52.
[41] Ahmed S., Sassatelli G., Torres L., Rouge L. ‘Survey of new trends in industry
for programmable hardware: FPGAs, MPPAs, MPSoCs, structured ASICs,
eFPGAs and new wave of innovation in FPGAs’. Proceedings of the 20th
International Conference on Field Programmable Logic and Applications;
Milano, Italy, 2 Sep; 2010.
[42] Glasser M. Open verification methodology cookbook, springer-­verlag new
york 2009, ISBN: 978-­1-­4899-­8513-­2; 2009.
[43] Kuon I., Rose J. ‘Measuring the gap between FPGAs and ASICs’. IEEE
Transactions on Computer-­Aided Design of Integrated Circuits and Systems.
2007, vol. 26(2), pp. 203–15.
[44] Monmasson E., Cirstea M.N. ‘FPGA design methodology for industrial con-
trol systems—a review’. IEEE Transactions on Industrial Electronics. 2007,
vol. 54(4), pp. 1824–42.
[45] Singh L., Drucker L., Khan N. Advanced verification techniques: a systemC
based approach for successful tapeout. Academic Publisher Kluwer; 2004.
 An intelligent verification management approach  59

[46] McKinney M.D. ‘Integrating Perl, Tcl and C++ into simulation-­based ASIC
verification environments’. Sixth IEEE International High-­Level Design
Validation and Test Workshop; Monterey, CA, 9 Nov; 2001. pp. 19–24.
[47] Haidar M., Kumar S. ‘Smart healthcare system for biomedical and health care
applications using aadhaar and blockchain’. 5th International Conference
on Information Systems and Computer Networks, ISCON 2021, 2021; GLA
Mathura, 22–23 Oct 2021; 2022. pp. 1–5.
[48] Kumar S., Cengiz K., Vimal S., Suresh A, et al. ‘Energy efficient resource
migration based load balance mechanism for high traffic applications iot’.
Wireless Personal Communications, 1-­14. 2021, vol. 10(3).
[49] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[50] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[51] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus network-­on-­chip architecture’. International Journal of Innovative
Technology and Exploring Engineering (IJITEE), 1672-­1676. 2019, vol. 8(6).
[52] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[53] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[54] Singh P., Bansal A., Kamal A.E., Kumar S., Marla D.. (eds.) ‘Road surface
quality monitoring using machine learning algorithm’ in Reddy A.N.R.,
Favorskaya M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and
Energy Sustainability. Smart Innovation, Systems and Technologies. 265.
Singapore: Springer; 2022.
[55] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and iot for
smart road traffic management system’. IEEE India Council International
Subsections Conference, INDISCON; Visakhapatnam, India, IEEE, 2020. pp.
289–96.
[56] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation
of fault tolerance technique for internet of things (iot)’. 12th International
Conference on Computational Intelligence and Communication Networks
(CICN); Bhimtal, India, IEEE, 2020. pp. 154–59.
[57] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. 10th International
Conference on Cloud Computing, Data Science and Engineering; Noida,
India, IEEE, 2020. pp. 63–76.
60  Intelligent network design driven by Big Data analytics

[58] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to

survive a disaster’. 4th International Conference on Information Systems and
Computer Networks, ISCON; Mathura, India, IEEE, 2019. pp. 250–55.
[59] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
International Conference on Computational Intelligence and Communication
Networks, CICN; IEEE, 2015. pp. 79–84.
[60] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium, Publisher: The Electromagnetics
Academy, PIERS 2015, 6-­9 July 2015, Prague, Czech Republic.; 2015. pp.
2363–67.
[61] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wireless
sensor network’. 5th International Conference on Communication Systems
and Network Technologies (CSNT); Gwalior, India, IEEE, 2015. pp. 194–200.
[62] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, IEEE, 2014.
[63] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems; Singapore: Springer; 2020.
[64] Kumar S., Trivedi M.C., Ranjan P., Punhani A. Evolution of Software-­Defined
Networking Foundations for IoT and 5G Mobile Networks. 10. Hershey, PA:
IGI Publisher; 2020. p. 350.
[65] Sampathkumar A., Rastogi R., Arukonda S., Shankar A., Kautish S., Sivaram
M. ‘An efficient hybrid methodology for detection of cancer-­causing gene
using CSC for micro array data’. Journal of Ambient Intelligence and
Humanized Computing. 2020, vol. 11(11), pp. 4743–51.
[66] Nie X., Fan T., Wang B., Li Z., Shankar A., Manickam A. ‘Big data analyt-
ics and IoT in operation safety management in under water management’.
Computer Communications. 2020, vol. 154, pp. 188–96.
[67] Shankar A., Jaisankar N., Khan M.S., Patan R., Balamurugan B. ‘Hybrid
model for security‐aware cluster head selection in wireless sensor networks’.
IET Wireless Sensor Systems. 2019, vol. 9(2) 68–76.
 Chapter 4
Evaluation of machine learning algorithms
on academic big dataset by using feature
selection techniques
Mukesh Kumar 1, Amar Jeet Singh 1, Bhisham Sharma 2, and
Korhan Cengiz 3

Identifying the most accurate methods for forecasting students’ academic achieve-
ment is the focus of this research. Globally, all educational institutions are con-
cerned about student attrition. The goal of all educational institutions is to increase
the student’s retention and graduation rates and this is only possible if at-­risk stu-
dents are identified early. Due to inherent classifier constraints and the incorporation
of fewer student features, most commonly used prediction models are inefficient
and incur. Different data mining algorithms like classification, clustering, regres-
sion, and association rule mining are used to uncover hidden patterns and relevant
information in student performance big datasets in academics. Naïve Bayes, random
forest, decision tree, multilayer perceptron (MLP), decision table (DT), JRip, and
logistic regression (LR) are some of the data mining techniques that can be applied.
A student’s academic performance big dataset comprises many features, none of
which are relevant or play a significant role in the mining process. So, features with
a variance close to 0 are removed from the student’s academic performance big data-
set because they have no impact on the mining process. To determine the influence
of various attributes on the class level, various feature selection (FS) techniques
such as the correlation attribute evaluator (CAE), information gain attribute evalu-
ator (IGAE), and gain ratio attribute evaluator (GRAE) are utilized. In this study,
authors have investigated the performance of various data mining algorithms on
the big dataset, as well as the effectiveness of various FS techniques. In conclu-
sion, each classification algorithm that is built with some FS methods improves the
performance of the classification algorithms in their overall predictive performance.

1
Department of Computer Science, Himachal Pradesh University, Summer-­Hill Shimla-­5, Himachal
Pradesh, India
2
Chitkara University School of Engineering and Technology, Chitkara University, Himachal Pradesh,
India
3
Department of Electrical-­Electronics Engineering, Trakya University, 22030 Edirne, Turkey
62  Intelligent network design driven by Big Data analytics

4.1   Introduction

Today, universities are in a very competitive and complicated situation. There has
been a lot of progress in technology and IT equipment, which has made it easier to
store a lot of data in educational databases. If these data are not analyzed, it is just
a lot of data. These data can be used with tools, methods, and techniques that help
us look at it. We can look for patterns and hidden information. Data mining is a way
to look for patterns and relationships in data that can help people make better deci-
sions. People from different fields work together in this area. It combines techniques
from statistics and artificial intelligence with those from neural networks, database
systems, and machine learning.
Data mining offers a variety of methods for analyzing data. A human’s ability
to assess and extract the most important information from student databases is cur-
rently limited by the sheer volume of data available. It is the extraction of nontrivial,
unknown, and potentially relevant information from a huge database through the
process of knowledge discovery. A user’s needs are taken into consideration when
data mining is employed in knowledge discovery. A linguistic term for describ-
ing a subset of data is referred to as a “pattern definition.” Data mining has a wide
range of applications. Financial institutions employ it to uncover hidden correla-
tions among various financial indicators, which they then use to spot potentially
fraudulent activity. It is able to identify both fraudulent and legitimate behavior by
analyzing previous data and turning it into valuable and accurate information [1].
By using data mining in healthcare, it is possible to identify correlations between
disease and treatment success. It also aids health-­care insurance providers in detect-
ing fraud. This software is often used to find money laundering, drug trafficking,
and other crimes. Customer demographic features and predicted behavior are two
frequent uses of data mining in the telecommunications industry to boost profit-
ability as well as reduce customer turnover. Based on the findings of data mining,
marketing campaigns and pricing strategies can be developed. Data mining tech-
niques are used in sales and marketing to uncover previously unnoticed patterns in
past customer purchases. In market basket analysis, the results of data mining can be
utilized to determine client purchasing habits and behavior patterns. Future trends
and customer purchasing habits can be predicted using these data. When it comes to
predicting how many customers will leave, the banking industry relies a lot on data
mining. The increasing use of technology in educational systems has resulted in an
enormous amount of data being made available to educators. A substantial amount
of important information may be gleaned via educational data mining (EDM), which
helps to provide a more accurate picture of learners and their learning processes.
This software analyzes educational data and helps students overcome educational
problems using data mining techniques. Similarly, to other data mining techniques’
extraction processes, EDM extracts information from educational data that are
engaging, interpretable, valuable, and innovative for the learner or teacher. EDM,
on the other hand, is primarily geared toward the development of methods that make
use of data that are unique to educational systems. After that, such strategies are
 Evaluation of machine learning algorithms  63

used to increase understanding of educational phenomena and pupils, as well as the

environments in which they learn [2]. Data and theory can help improve the quality
of teaching and learning in schools by developing new computational methods that
combine data and theory.
When it comes to the practical use of EDM, users can extract knowledge from stu-
dent data. There are numerous applications for this knowledge, including the validation
and evaluation of educational systems, the improvement of teaching and learning pro-
cesses, and the establishment of the groundwork for a more effective learning process,
among other things. Comparable concepts have been used successfully, particularly in
commercial data, in many datasets, like online shopping carts, to boost sales profitabil-
ity. Because data mining techniques have been proven successful in commercial data,
they are being used more widely in a variety of fields. Notably, data mining has been
applied to educational data in order to achieve research objectives such as improving
the learning process and guiding students’ learning, as well as gaining a more compre-
hensive understanding of educational phenomena. The situation is changing, however,
as more people become interested in the application of data mining in the educational
environment. Even though EDM has made less progress than other industries in this
area, things are getting better.
Nowadays, universities operate in a very difficult and competitive atmosphere.
It’s becoming more and more common for educational databases to contain vast
volumes of information, yet these data are useless if they are not analyzed further.
It’s possible to analyze these data using data mining tools, methods, techniques,
and procedures. To determine the influence of various attributes on the class level,
various FS techniques such as the CAE, IGAE, and GRAE are utilized. A crucial
stage in the development of any country, and indeed in the development of our entire
planet, is the acquisition of education. Consequently, data pertaining to the educa-
tion sector is thoroughly analyzed in order to extract some useful information from it
that will aid in the further development of students in the future. Consequently, data
mining and machine learning are of paramount relevance for this purpose because
there are numerous ways available to aid in the discovery of relevant information
inside a large database of this nature [3].
Classification, clustering, and association rule mining are just a few of the data
mining techniques that can be utilized to find the information that’s been concealed
in a database. Machine learning, together with data mining, is one of the most effi-
cient and successful methodologies for forecasting performance in the disciplines of
data analytics and information retrieval, respectively. It is used to forecast perfor-
mance by constructing models and algorithms, similar to how data mining is used.
Machine learning is used to learn and act in ways that are similar to those of human
beings. Machine learning algorithms are often classified by either learning style or
similarity in form or function, for example, classification, regression, decision tree,
clustering, and deep learning algorithms, to name a few examples. A large number
of attributes exist in the real-­world database, and none of these attributes are neces-
sary for predicting the pattern in a dataset to be found there. As a result, we must
delete some features from the database in order to speed up the mining process and
produce more accurate results. As a result, we have data mining techniques that may
64  Intelligent network design driven by Big Data analytics

be used to remove certain attributes from a database for this purpose. FS is an impor-
tant approach in the success of the data mining process since it allows us to select
the helpful or relevant qualities in the dataset that are being used for the analysis [4].
In this study, a comparison of alternative FS approaches is made, as well as an
examination of the impact of the strategies on classification algorithms. Any coun-
try’s development depends on its educational system. So, take it seriously from the
outset. They have their own systems and evaluators. Today’s education includes
online education, Massive Open Online Courses (MOOC) courses, intelligent tutori-
als, web-­based education, project-­based learning, seminars, and workshops [5]. But
none of these systems work unless they are accurately evaluated. So, to make any
school system successful, a clear evaluation mechanism is needed. Every educa-
tional institution creates a lot of data about each registered student, and if that data
are not correctly analyzed, all resources are lost. It includes student acceptance, fam-
ily info, and academic results. Every educational institution assesses their students
[6]. In modern education, many tools are utilized to measure a student’s academic
achievement. Data mining is one of the greatest sophisticated computer tools used
to monitor student progress. Currently, data mining is applied in almost every field
where data are used. Data mining has several key applications in retail, marketing,
banking, telecommunications, hospitality, hospital, and production management.
This entire company uses data mining to increase sales and future growth. With the
help of a built-­in algorithm, it analyzes any organization’s historical data to uncover
hidden information [7]. So, we may say that data mining techniques can extract hid-
den information from any organization’s data warehouse.

4.1.1   EDM
In Montreal, EDM hosted its first international research conference, which took
place in 2008. Founded in 2011, the International Educational Data Mining Society
is dedicated to advancing EDM. Since then, disciplines such as data mining, machine
learning, pattern recognition, psychometrics, artificial intelligence, information vis-
ualization, and computational modeling have increased in popularity. EDM’s ulti-
mate goal is to improve education and to make decision-­making processes more
transparent [8]. Data that are made during teaching and learning can be used to find
new information, correlations, and patterns in very large datasets.
Data mining has a wide range of applications. Customer behavior data may be
used to improve customer loyalty, and it can also be used to uncover hidden correla-
tions between various financial indicators in order to spot potentially illegal activity.
It collects and analyzes past data in order to detect both fraudulent and nonfraudulent
behavior. Data mining in healthcare makes discovering the connections between dis-
eases and therapies easier [9]. Fraud detection is made easier for health-­care insurers
as well. It is used by law enforcement authorities to investigate money laundering,
narcotics trafficking, and other criminal activities. Customer demographic features
and predicted behavior are two frequent uses of data mining in the telecommuni-
cations industry to boost profitability as well as reduce customer turnover. Based
on the findings of data mining, marketing campaigns and pricing strategies can be
 Evaluation of machine learning algorithms  65

developed. Using data mining tools, companies can uncover previously unnoticed
trends in customer purchase behavior [10]. Data mining results are used to discover
customers’ purchasing habits by analyzing market baskets and analyzing the combi-
nations of products they purchase together. Additionally, it’s used to make educated
guesses about upcoming fashion and consumer trends. A significant amount of data
mining is used in the banking business when projecting how many clients will quit.
The amount of data produced and saved in many educational institutions has grown
to the point that it is impossible to analyze it manually any longer. A new subject,
EDM, has emerged as a result of the analysis of educational data.

4.1.2   EDM process

The EDM procedure is divided into four distinct stages. There are two stages to
solving a data mining challenge: problem definition and problem translation. Along
with the primary research topics, the project’s goals and objectives are developed
during this stage. It takes the most time to prepare and acquire data in the second
phase. It can account for up to 80% of the total time spent on the investigation. In
data mining, the quality of the data is a crucial issue. Source data must be discov-
ered, cleansed, and formatted according to predetermined specifications in this step
[11]. Ultimately, the parameters are set to their greatest potential values and vari-
ous modeling techniques are applied. During the final stage of data mining, known
as deployment, the results are organized and presented graphically and in reports.
Because data mining is a recursive process, you must remember that it does not end
when a specific solution is applied. It only has to be incorporated into a brand-­new
data mining workflow. To properly analyze the database, some important and fre-
quently used data mining techniques are applied to get hidden information. Some of
the major advantages of data mining process are mentioned below:

1. Data mining is used to discover patterns, forecast trends, and find relations, thus
enabling the companies to decode the true potential of their precious data. The
game-­changing insights derived from data mining process enable the organiza-
tions to make informed business decisions, devise effective strategies, and gain
a competitive advantage in the industry.
2. Organizations can learn in detail about the factors affecting and impacting mar-
ket demographics. Alongside, they can uncover new business prospects, tap
untouched markets that might extend beyond geographical boundaries, and
expand the horizons.
3. They can understand the consumer better, know their preferences, age, spend-
ing habits, social media behavior, which product is preferred the most, etc.,
and offer them customized services accordingly. This consequently leads to
increased loyalty and lesser churned out rate.

Data mining is described as the procedure of extracting hidden information from

any larger raw dataset. It means with one or more software; the data patterns can be
examined in huge data. It has applications in many areas, like science and research.
66  Intelligent network design driven by Big Data analytics

As a data mining application, companies can adapt more about their clients and can
execute more efficient schemes for a number of business functions to leverage the
resources in a more optimized and perceptive manner. This assists the companies to
get closer to the goals and make better results.
It involves efficient data compilation, warehousing, and computer processing.
To segment the data and to estimate the possibility of future events, it utilizes com-
posite mathematical algorithms. It is also termed as knowledge discovery in data-
bases (KDD). The main features of data mining are automatic pattern predictions
based on trends, behavior analysis. The focus is given to huge datasets and data-
bases, for discovering and documenting the fact groups based on clustering that are
not known previously.
The algorithms of data mining are the step-­by-­step definition of the process
utilized to bring meaning to the set of information. Some of them are quite simple
and need little to comprehend and execute and some are very composite and need
significant learning and an attempt to implement. The algorithms could be of several
forms that depend on the data to be undertaken and the outcome to attain. In this
section, numbers of data mining algorithms are defined with its types.

4.1.3   Methods and techniques

In the field of EDM, a diverse range of approaches and procedures are used. As a
general rule, categorization and classification are the most often encountered appli-
cations. Among the most widely used methodologies are neural networks, decision
trees, regression, and cluster analysis, to name a few examples.
Classification algorithms: In general, classification is a data mining technique
that classifies data into target categories or classes based on the frequency with
which they exist in a collection of information. With the help of this tool, analyzing
and forecasting results become much easier. Classification must be done on each
instance in order to achieve accurate prediction of the target class for each case in
the dataset as a whole. The classifier training algorithm makes use of preclassified
examples in order to identify the set of parameters that must be considered for clas-
sification [12]. This technique, which is frequently used in the educational sector,
is used to classify students according to certain characteristics such as their age and
gender, their grades and other academic achievements, their motivation and behav-
ior, as well as their demographic and geographic characteristics.
Clustering algorithms: Clustering analysis is a technique for generating
new clusters from datasets that have already been established. During the data
preprocessing step, it is beneficial to identify homogeneous groups that can be
used as input into subsequent models. A cluster analysis can be used to analyze
and contrast different types of things such as students, courses, instructors, and
other similar entities, among others. In the field of prediction, the practice of
making calculated assumptions about future events on the basis of evidence that
has been thoroughly investigated is defined [13]. It is possible to forecast the rela-
tionship between one or more independent variables and one or more dependent
variables using the regression method. When conducting research, it is critical
 Evaluation of machine learning algorithms  67

to distinguish between independent variables and response variables as much

as possible. A wide range of applications, including business planning, finan-
cial forecasting, time series prediction, and trend analysis, can be made using
this technique. Predictive analytics is used in the educational business to produce
predictions regarding students' academic performance [14]. Examples include
enrollment forecasts, final grade predictions, dropout predictions, and other simi-
lar predictions.
Association rule mining algorithm: Analyzing the likelihood of elements in a
collection being associated is a data mining approach known as “association.” The
rules that control the relationships between things that occur in close proximity to
one another are known as “association rules.” Most of the time, when it comes to
the analysis of sales transactions, association rules are applied. In addition to sales
promotions and direct marketing, catalog design, cross-­sell marketing, and recog-
nizing company trends are also possible applications for this type of finding [15].
This method, which is regulated by a set of rules, can be used to start new courses or
to start new educational institutions.
There are a lot of benefits and uses for EDM that will be talked about in more
detail below. Recent years have seen the publication of a number of research papers
and studies that look at how data mining can be used in educational settings. It is
planned to go into deeper detail on some of these eventually. There are a lot of ways
EDM can be used. For example, it can be used to help make decisions about which
courses students should take. It can also be used to make better predictions about
how students will do.
K-­nearest neighbor (KNN): KNN is a nonparametric approach for classifying
data points based on their proximity and correlation with other accessible data. This
technique makes the assumption that similar data points can be located in close
proximity to one another. As a result, it attempts to determine the distance between
data points, typically using Euclidean distance, and then assigns a category based on
the most frequent or average category.
Neural networks: Often used in conjunction with deep learning algorithms,
neural networks process training data by simulating the interconnectedness of the
human brain via layers of nodes. Each node is composed of several components:
inputs, weights, a bias (or threshold), and an output. If the output value exceeds
a predefined threshold, the node “fires,” or activates, transmitting data to the net-
work’s next layer. Neural networks acquire this mapping function by supervised
learning, changing according to the loss function via the gradient descent process.
When the cost function is close to or equal to zero, we can be confident that the
model will produce the correct answer.
In the rest of the chapter, the Materials and Methods section discusses the
data preprocessing section on the selected datasets. The proposed algorithm and
its implementation are presented in Section 4. The Results and Discussion section
includes the result and discussion of this study. Finally, the Conclusion section
concludes the article, and future scope directions are highlighted.
68  Intelligent network design driven by Big Data analytics

4.1.4   Application areas of data mining

Data mining is frequently employed in a wide variety of fields. There are numerous
commercial data mining systems available today, but this field faces numerous prob-
lems. Modern businesses cannot exist in a data lacuna. They must evolve and adapt
to technological advancements and emerging digital trends in order to stay ahead of
the competition. As a result, organizations now place a premium on remaining cur-
rent with all advancements in the field of data science and analytics. Data mining is
one of these data science processes. It entails the analysis of already created datasets
in order to glean new and relevant information. By segmenting massive datasets,
recognizing trends, and forecasting outcomes, complicated data mining algorithms
enable businesses to make sense of raw data. Data mining techniques are frequently
used by business intelligence and data analytic teams to aid in the extraction of
knowledge specific to their organization and industry. Several data mining applica-
tions include the following.
Data mining in healthcare: The application of data mining in the health-­care
field has resulted in some outstanding outcomes. With the use of data mining, it
is possible to analyze and discover some of the health-­care practices that serve to
ensure quality while also assisting in cost reduction.
Sales and marketing: Companies acquire a vast quantity of data about their cus-
tomers and prospects. By observing consumer demographics and online user activ-
ity, organizations may use data to optimize their marketing campaigns, improving
segmentation, cross-­sell offers, and customer loyalty programs, generating higher
return on investment (ROI) on marketing efforts. Predictive analytics can also allow
teams to set expectations with their stakeholders, offering yield projections from any
increases or cutbacks in marketing investment.
Data mining in fraud detection: The procedures employed to detect fraudulent
activities proved to be time-­consuming. Detecting fraud has gotten simpler as a
result of the emergence of data mining techniques. With the use of data mining, it
has become easier to spot trends in data and to take steps to protect the privacy of
users’ information.
Data mining in customer segmentation: It assists in the segmentation of custom-
ers into groups so that products and services can be tailored to meet their specific
demands.
Operational optimization: Using data mining techniques, process mining
reduces operational expenses, allowing firms to run more effectively. A corporate
leader’s decision-­making process has been improved by this method, which identi-
fies costly obstacles.
Data mining in banking and finance: With the advent of broad computerization
in the sector of banking, it has become easier to do data analysis and interpretation.
When it comes to company information, it may be utilized to determine the relation-
ship between information about the firm and the prices that are prominent in the
market.
Data mining in multimedia: It is one of the most recent ways that is catching
on because of the increasing capacity to extract important data with high accuracy.
 Evaluation of machine learning algorithms  69

In this process, data are extracted from several types of multimedia sources, such
as audio, text and hypertext, video and still images, and the data are turned into a
numerical representation that may be shown in a variety of forms, such as HTML.
This method can be used for a variety of tasks, such as grouping and classification,
performing similarity checks, and identifying relationships.
Data mining in distributed architecture: It is becoming increasingly popular
because it allows for the mining of large amounts of information that is kept in mul-
tiple locations within a firm or across multiple companies. In order to collect data
from many sources and deliver appropriate insights and reports based on it, highly
sophisticated algorithms are used.
Data mining in education: The application of data mining in the realm of edu-
cation has resulted in the development of what is known as EDM. It aids in the
prediction of students’ learning tendencies by providing them with data. It also aids
in the analysis of learning outcomes and the formulation of decisions based on the
information gained. According to the education data mining community, it is a new
discipline that focuses on developing techniques to investigate unique data types
from the educational environment and using them to better understand students and
learn from them.
It is becoming increasingly popular to use data mining technologies in the
subject of education, as seen by the tremendous increase in interest. Development
approaches for discovering information from data obtained from the educational
environment are part of this growing topic, which is now under investigation. This
technique differs from standard data mining techniques in that it explicitly employs
many degrees of meaningful hierarchy in educational data, as opposed to traditional
data mining techniques. EDM is concerned with the collecting, archiving, and anal-
ysis of data that are connected to student learning and evaluation in a formal setting.
It is common for researchers conducting EDM studies to use approaches that have
been pulled from a variety of sources, such as psychometrics, machine learning, data
mining, educational statistics, information visualization, and computer modeling.
An interactive loop is created by the implementation of data mining in educational
systems, which includes the phases of formation, testing, and improvement. The
knowledge that has been discovered should be entered into the system’s cycle, and
the guide should support and boost overall learning. The system is used to transform
data into knowledge and to filter the knowledge obtained through mining in order to
aid in decision-­making processes.

4.2   Literature survey

To complete this study on predicting academic performance of the students using dif-
ferent data mining techniques, we came across different research and review papers.
During literature study on this topic, we found that a most of the researcher of EDM
communities are trying to develop a system which effectively and efficiently predict
the academic performance of the students. Although a lot of data mining algorithms
are development in the past like classification algorithms that effectively predict the
70  Intelligent network design driven by Big Data analytics

class level of any datapoint in the data. Most of the researcher in the literature con-
sidered different aspects of student attributes in consideration while collecting data
for the dataset. We have different categories of student’s attributes like academic,
personal, family, social and institutional, and all these attributes contribute in the
development of predictive models for academic performance prediction [16]. EDM
is becoming more popular nowadays as a result of the proliferation of electronic
resources, the use of online educational tools, and the Internet. Numerous studies are
being conducted to improve educational materials and technologies. By leveraging
EDM techniques to forecast or analyze student performance and to assist students
who are receiving less than satisfactory grades, an artificial neural network classifier
model was created that can benefit both students and teachers in discovering knowl-
edge from the massive amount of data available in the educational sector.
Sentiment analysis was used to better understand students’ learning styles and
study plans in order to improve instruction [17]. After removing behavioral infor-
mation from students, a model was suggested using data mining approaches that
achieved a 22.1% high accuracy after removing behavioral features. Additionally, it
was discovered that applying ensemble methods resulted in a 25.8% improvement
in accuracy. The academic dataset contained 473 cases, and the Bayesian classifier
achieved a 70% accuracy rate. To characterize student dropouts, the Naïve Bayes’
classifier, artificial neural network (ANN), KNN, and J48 were employed. KNN
and decision trees with 10-­fold cross-­validation achieved 87% and 79.7% accuracy,
respectively. By building a hyperplane, support-­vector machines (SVMs) distin-
guish classes in high-­dimensional space [18]. On health-­related data, data mining
techniques such as LR and multi-­classifiers achieved amazing results. Decision
trees employ optimal ways to identify or reach a specific aim in the real world.
Combining this principle with the ability to accelerate the training process explains
their extensive use in EDM. On a dataset of 15,150 instances, a decision tree was
utilized to predict whether courses pass or fail; 85.92% accuracy was achieved [19].
Another most important algorithms are there in Data mining that help to increase the
accuracy level of any data mining classification algorithms. We have FS techniques
like filter and wrapper method that help to reduce the different attributes of any
dataset that are not very helpful in making prediction on any class. So, in literature,
we came across many FS techniques like CAE (CorrelationAttributeEval), IGAE
(informationGainAttributeEval), and GRAE (GainRatioAttributeEval) with Ranker
as search method and are mostly used in literature [20].
During literature review we focus our review toward two main features: first
is to check that which attributes of the student mostly affect the academic perfor-
mance of the students and second which classification algorithms are mostly used by
researcher to predict the result or performance of the student in academics. The last
and most important point is to check the importance of FS in the prediction of the
academic result. And in most of the research, we found that FS improves the perfor-
mance of any prediction model. According to Hung et al., classification algorithms
like SVM, RF, and neural networks can all be used to identify at-­risk students. The
new method was better than other methods at both accuracy and sensitivity in tests
on two different datasets: one from a school and the other from a university [21]. The
 Evaluation of machine learning algorithms  71

K-­means algorithm was used to study the challenge of detecting the amount of stu-
dent participation in a similar way. Furthermore, using the a priori association rules
method, the authors were able to construct a set of rulers that were associated with
student involvement and academic success. Researchers discovered that students’
levels of involvement and academic performance in an e-­learning environment have
a favorable relationship, according to the results of their experiments.
According to Helal et al., students’ socio-­demographic characteristics as well as
their university admission basis and attendance type can be taken into account when pre-
dicting their performance. Rule- and tree-­based algorithms were shown to be the most
interpretable by the authors’ experiments, making them the most suitable for educational
settings [22]. Zupanc and Bosnic added semantic coherence and consistency character-
istics to an existing automated essay evaluation system. The authors demonstrated that
their proposed approach gave superior semantic feedback to the writer through experi-
mentation. Aside from that, the accuracy of its grading was superior to that of other
cutting-­edge automated essay evaluation systems. Two layers of machine learning were
proposed by Xu et al. to track and predict student performance in degree programs.
According to their simulations, the proposed strategy outperformed benchmark strate-
gies [23]. In this paper, the author compared the accuracy, F-­measure, and true positive
rate of 11 machine learning models. For the measures listed above, they found that the
decision tree method outperformed alternative classifiers. The number of scholars work-
ing in EDM has skyrocketed in the last several years.
It is possible to employ EDM in a wide range of applications, including user group-
ing and student performance prediction [24]. In order to better comprehend students,
teachers, and their many attributes, academics are developing new statistical methodolo-
gies and machine learning algorithms. Data analysis to improve student and teacher per-
formance is nothing new. As a result of recent developments in educational institutions,
such as the increase in computing facilities and the introduction of web-­based learning,
researchers are developing new ways for analyzing educational data. The information
can be obtained via admissions, registration, library management, syllabus, and course
management systems. It is possible to forecast student performance based on data, and
this study employs four machine learning methods to accomplish so: the decision tree,
the random forest classifier, gradient boosting classifier, and extreme gradient boosting
classifier, among others. In order to forecast student performance, researchers use a vari-
ety of categorization techniques [25].
The author of this paper also used decision tree techniques, neural networks, and
linear discriminant analysis to identify students who were at risk of failing the course
and divide them into three categories: low-, intermediate-, and high-­risk students [26].
Using classification techniques, it was possible to develop a profile of students who
were most likely to fail or succeed in their studies. In order to forecast students’ achieve-
ment, the author looked at socio-­demographic and educational data from the pupils.
Researchers used the Chi-­square Automatic Interaction Detector decision tree technique
to construct a predictive model that can identify students who are struggling to acquire
new material [27]. Use of the Naive Bayes (NB) classification by Pandey allowed him
to accurately identify bright kids from those who were more mediocre. Based on their
previous results, their program was able to accurately forecast students’ final grades.
72  Intelligent network design driven by Big Data analytics

A comparative study was undertaken in 2012 in order to estimate the student’s per-
formance. Study participants were asked to choose from among several decision-­tree
algorithms to see which one best predicted students’ grade [28]. After testing a number
of different classifiers, they came to the conclusion that a classifier’s precision and accu-
racy should be taken into consideration when deciding which one should be used. They
found that the Classification and Regression Tree algorithm (CART) algorithm, which
was made to look like the decision tree algorithm, was the best. Five machine learn-
ing categorization models are tested by Sekeroglu et al. to see which one best predicts
student achievement in higher education. According to the researchers’ findings, data
preprocessing procedures boost prediction accuracy [29].
So, after review, we selected an academic dataset from the University of California
Irvine (UCI) repository and tried to implement the selected classification algorithms on
that. Now, in our next step, FS algorithms are also used on the selected dataset and select
the top 10 attributes for our implementation with all the classification algorithms [30].
Romero and Ventura highlighted the following applications for EDM: student model-
ing, performance prediction, data visualization, social network analysis, feedback to help
managers, planning and scheduling, grouping students, and identification of undesirable
behaviors. To identify kids who are at risk of failing their first year of education, Pal
employed the categorization technique to analyze data from prior years’ student dropout
statistics [31]. Decision tree approaches were employed in conjunction with information
on the student’s previous education, family income, and parents’ education to forecast
the list of students that required further attention in order to lower the drop-­out rate. The
results of this study show that the machine learning model can use current student drop-
out data to build a good prediction model [32]. A deeper knowledge of student behavior
and learning styles is essential for academic institutions to effectively manage current
study programs and educational practices.
Data mining techniques and models can be used as decision-­support tools in
education by reviewing datasets and establishing the significance of the influence of
specific variables, resulting in more successful research and an overall improvement
in educational quality [33]. As a result of EDM, colleges can do a better job of allo-
cating resources to their students. Different data mining algorithms such as decision
tree, link analysis, and decision forest were utilized to investigate the preferences of
students for courses, as well as their completion rates and professions of enrollment
[34]. It was discovered through this study that there is a relationship between course
category and enrollment professions, and it was also discovered that data mining is
important for curriculum development and marketing in the sphere of higher edu-
cation These findings may be utilized as a guideline for marketing and curriculum
development efforts in the future.

4.3   Materials and methods

4.3.1   Dataset description

Students from two Portuguese secondary schools were studied for this dataset.
Students’ demographic, social, and school-­related information were gathered through
 Evaluation of machine learning algorithms  73

school reports and questionnaires. There is a math (mat) dataset and a Portuguese
language performance (pl) dataset (por). G3 is strongly linked to G2 and G1 objec-
tive attributes. Graduation is divided into three periods: G1, G2, and G3, with the
latter two representing the first and second semesters, respectively. Predicting the
outcome of G3 on its own is more difficult, but it is also more advantageous.

4.3.2   Classification algorithms

Classification is a data mining technique for sorting and classifying items in a collec-
tion. Each instance of data must be classified in order to accurately forecast its place
in the target class. A categorization model, for example, might be used to divide loan
applicants into three groups based on their credit risk: low, medium, and high. The
following are a few of the classification techniques that have been chosen for use:
Naïve Bayes’ algorithm: The Bayes’ theorem can be used to solve classification
issues using the Naïve Bayes’ technique, a supervised learning process. In order to
develop machine learning models that are accurate in a range of situations, this classifier
offers a simple and effective classification technique [35]. It is possible to use the Naïve
Bayesian classifier to categorize datasets and to classify datasets using Bayes’ theorem
and the assumption that the predictors are independent. It is simple to build a Naïve
Bayesian model since it does not require complicated iterative parameter estimations.
For huge datasets in particular, it is a success. Classifiers that are more complicated are
often outperformed by the Naïve Bayesian classifier because of its simplicity. Since it
routinely beats more complicated categorization algorithms in a wide range of situa-
tions, it has become an increasingly popular tool. In order to arrive at the posterior prob-
ability, P(c|x), we can use the Bayes’ theorem to combine the prior probabilities: P(x)
and P(x|c). Based on Naïve Bayes’ classification, a classifier (c) is decided by a predictor
(x) value that has no bearing on the other predictors’ values. The scientific literature
refers to this as class conditional independence.
Random forest algorithm: Random forests (also known as random decision for-
ests) are an ensemble learning technique for classification and regression problems that
involves training a large number of decision trees in a distributed form. Random for-
ests are useful for classifying problems because they produce the classification that was
selected by the majority of trees in the forest during classification. It is possible to find
this well-­known machine learning method in many different configurations as part of
the supervised learning strategy. For both classification and regression issues in machine
learning, machine learning can be used [36]. When it comes to solving complicated
problems, it uses the notion of ensemble learning, which is the act of merging multiple
classifiers into one model. For the purpose of increasing the projected accuracy of a
dataset, the Random Forest classifier “combines a large number of decision trees on dif-
ferent subsets of that dataset and takes an average.” With the random forest, instead of
relying on a single decision tree, it takes into account the projections from each tree and
predicts the final outcome based on a majority of votes. When a forest contains an exces-
sive number of trees, overfitting might become an issue. Trees are more accurate and less
likely to appear in a forest with a large number of trees. Using a random forest to predict
the class of a dataset improves the likelihood that certain trees will predict the correct
74  Intelligent network design driven by Big Data analytics

output while other trees will not. All the trees, however, properly predict the correct out-
come when taken together. In order for the classifier to be able to accurately predict out-
comes rather than make educated guesses, the dataset’s feature variable must have a few
real values. The correlation between the forecasts of each tree must be extremely low.
Decision tree algorithm: In statistics, data mining, and machine learning, the
decision tree algorithm, also known as induction of decision trees, is a predictive
modeling tool. From the observed qualities of an object’s goal value, it employs
a decision tree [37]. Supervised learning approaches such as decision tree can be
used to address classification and regression problems; however, it is more typically
used to tackle classification difficulties. Each leaf node in a tree-­structured classi-
fier represents a classification result, with internal nodes relating to dataset features
and branching corresponding to rule sets. In a decision tree, the decision node and
the leaf node are two of the nodes that make up the tree’s branches. Instead of mak-
ing decisions, decision nodes have several branches, but leaf nodes only report the
results of those decisions. Decisions and tests are based on the dataset’s character-
istics. Graphic representations of all conceivable outcomes are used in this strategy.
This is why the term “decision tree” was coined: it expands outward in a tree-­like
form from the root node. The CART method, which stands for Classification and
Regression Tree algorithm, is used to create a tree for the purpose of classification
and regression. According to the answer, the tree is divided into subtrees in a deci-
sion tree, which is a sort of decision tree (yes or no).
MLP algorithm: The MLP differs from a linear perceptron in that it has multiple
layers and does not activate in a linear fashion. It has the capability of separating data
that are not linearly separable, among other things. MLPs are feedforward artificial neu-
ral networks that generate outputs from a collection of inputs. A directed graph com-
bines the input and output layers of an MLP’s input nodes together to form a single unit
known as a directed graph unit. Backpropagation is used by MLP to train the network
and improve its performance. MLP is a deep learning method that uses machine learning
[38]. Given that it is a neural network connected to several layers in a directed graph,
the signal flow through the nodes of an MLP is restricted to one direction only. There
are no linear activation functions in this system; instead, all nodes, with the exception
of the input nodes, have non-­linear activation functions as their activation functions. An
approach known as backpropagation is used by MLPs to learn from their mistakes. MLP
is classified as a deep learning technique since it makes use of multiple layers of neu-
rons. MLP is frequently utilized in research in computational neuroscience and parallel
distributed processing. It is also commonly used for supervised learning tasks, which is
why it is so popular. Speech and image recognition, as well as machine translation, are
just a few of the many applications available today.
DT algorithm: DTs are scheduled rule logic entries that are table-­formatted and
contain conditions (represented by the row and column names) and actions (represented
by the intersection points of the table’s conditional cases). DTs are used to organize and
schedule rule logic entries. DTs are particularly useful when dealing with business rules
that have a number of different conditions to consider. A decision tree or a switch-­case
statement in a programming language can also represent the information in a DT. A
DT, sometimes referred to as a cause–effect table, is an effective tool for determining
 Evaluation of machine learning algorithms  75

the relationships between various inputs and their associated outputs. The DT is derived
from a logical diagramming approach known as cause–effect graphing, which is why
the term “cause-­effect table” was coined. Developing tests are made considerably easier
by using DTs. Using this tool, testers can investigate the consequences of diverse inputs
and other software states on business rules that must be correctly implemented in soft-
ware [39]. When it comes to developing and testing complicated business rules, it gives
a consistent approach. Developers will be able to do a better job with the help of this
tool. Complex business rules necessitate the use of an organized approach to preparing
requirements. Complex logic is another use for it.
JRip algorithm: It is represented by this class as the implementation of a prop-
ositional rule learner. This method, which is an optimized variant of the IREP, was
devised by William W. Cohen. It is named after him. It performs repeated incremen-
tal pruning in order to reduce the number of errors (RIPPER). With its bottom-­up
approach to rule identification and learning, JRip classifies instances in the training
data into groups and then discovers the set of rules that apply to all members of each
group. Techniques such as cross-­validation and having a specific quantity of words
are employed in order to avoid overfitting the model.
Linear regression algorithm: Using LR, the odds ratio may be calculated
even when there are many explanatory variables present. A binomial distribution
rather than a linear distribution is used as the response variable in this method,
which is very similar to multiple linear regression. Observed occurrence odds
ratios are used to determine the outcome of the experiment. Linear regression, a
fundamental statistical regression approach, can be used to do predictive analysis
and explain the link between continuous variables in a calm and straightforward
way [40]. Linear regression is a statistical approach that develops linear relation-
ships between independent and dependent variables in accordance with its name.
As a result, linear regression is the name given to this technique. Simplified linear
regression refers to linear regression with only one input variable when there are
no other variables in the dataset. The term “multiple linear regression” refers to
the fact that the equation contains many different input variables. The output of
this model is a slanted straight line that depicts the relationship between the two
variables under consideration.

4.3.3 FS algorithms
Methods for reducing the number of variables used in a model’s prediction of a
target variable are known as FS. One way to think about an algorithm for selecting
new feature subsets is to think of it as an amalgamation of a search approach and
an evaluation measure. In order to identify the algorithm with the lowest error rate,
we may simply test all potential subsets of characteristics and see which one works
best. By reducing redundant and superfluous data, FS is a simple but effective solu-
tion to this problem [41]. A better knowledge of the model or data can be gained by
eliminating the irrelevant data, which increases accuracy and decreases computing
time. As a fundamental machine learning technique, FS helps focus the use of vari-
ables on those that are most useful and efficient for a given machine learning system.
76  Intelligent network design driven by Big Data analytics

Correlation attribute evaluator: The correlation attribute is used to compare and

contrast the attributes in respect to the target class, which is defined as follows: Pearson’s
correlation approach is used to determine the degree of correlation between each attribute
and the target class attribute in each attribute. It takes nominal properties into account on
a value basis, with each value acting as an indicator. The educational dataset is evaluated
using a combination of this attribute evaluator and the search ranking method, which is
applied in conjunction with each other [42]. A correlation coefficient (Pearson’s correla-
tion coefficient) between an attribute and a class is used to determine the value of the
attribute. Nominal attributes are analyzed value by value, with each value serving as a
signal of the next value in the analysis. When determining the overall correlation for a
nominal attribute, a weighted average is utilized (Figure 4.1).
Information gain attribute evaluator: The for-­ attribute’s information gain
is equal to the difference between the training set’s a priori Shannon entropy and
the conditional entropy. If the outcome attribute can be categorized individually for
each of the attribute values, mutual information is equal to the total entropy [43].
A reduction in entropy or surprise can be quantified by dividing a dataset according
to the value of a random variable. In this situation, a smaller entropy group or group
of samples and consequently less surprise are implied by a bigger information gain.
Information is a statistic for measuring how shocking an event is in terms of the num-
ber of bits it contains. According to the probability of an event, incidents that are less
likely to occur are better documented than those that are more likely [44]. In statistics,
the probability distribution of a random variable is referred to as its entropy, and it is a
measure of the amount of information it contains. An equal probability distribution, in

Figure 4.1  Graphical representation of accuracy level of classification

algorithms
 Evaluation of machine learning algorithms  77

contrast to a skewed distribution, contains more information because all events have the
same probability.
Gain ratio attribute evaluator: This method activates an evaluator for the gain
ratio attribute and determines the value of an individual attribute by calculating the
gain ratio of the class that contains the attribute. This method returns the evalua-
tor’s capabilities and allows an attribute evaluator to perform any additional post-­
processing on the given attribute collection [45].

4.3.4 Data preprocessing phase

Algorithm-­1 illustrates the main methods for filling in missing data and resolving the
problem of an imbalanced dataset, referred to as defects elimination utilizing the mean
and median processes and resolving the imbalance problem. To begin developing the
process model, this work performs data preprocessing at the lowest level. For data pre-
processing, a proposed algorithm is presented that addresses the dataset’s missing value
and imbalance issues.

‍ ‍

The targeted output class has a range of 0–20, and there are 21 clusters in the ini-
tial configuration. In terms of the classification task, this is an unreasonable solution
78  Intelligent network design driven by Big Data analytics

Table 4.1   New class level assigned to the dataset

Range of initial class given in

the dataset New cluster number assigned to the class level
Grade between 16 and 20 Class A
Grade between 14 and 15 Class B
Grade between 12 and 13 Class C
Grade between 10 and 11 Class D
Grade less or equal to 9 Class F

because it makes the classification process incredibly difficult, especially given the
limited number of instances provided. In the given dataset, G1, G2, and G3 and the
grade obtained by different students and for better results we find the final grade of
the student by finding the average of all grades and create a new attribute named
“Total Grade.” As a result, I’ve assigned a group of clusters to a few class levels
denoted by the letters A, B, C, D, and F in Table 4.1.

4.4 Implementation of the proposed algorithms

To identify the best classifiers that accurately generalized the data, we implemented
various classification algorithms on a dataset of student academic performance. We
chose different parameters for the algorithms that effectively analyze the dataset
and increase their generalization accuracy [46]. Please keep in mind that all imple-
mented classification algorithms undergo a 10-­fold cross-­validation to determine
their correctness.

4.4.1 Model construction for the standard classifier

Numerous classification techniques were chosen and used to the dataset of student
performance. We implement the following classification algorithms: Naïve Bayes,
random forest, J48, MLP, DT, JRip, and LR. The table below summarizes the imple-
mentation results of various categorization algorithms using ten cross-­validation
(k-­fold cross-­validation) approaches. As our dataset is a balanced dataset with nearly
equal distribution of data across five distinct classifications. According to Table 4.2,
decision tree classification method had the greatest accuracy of 96.76% when com-
pared to other classification algorithms such as Naïve Bayes, random forest, DT,
MLP, JRip, and LR. As shown, the MLP method achieves the lowest accuracy of
84.59%. Random forest and JRip algorithms also obtained an acceptable level of
accuracy, at 92.14% and 96.14%, respectively. To implement these algorithms, all of
the dataset’s attributes (up to 32) are considered. Other performance metrics such as
mean absolute error (MAE), precision, recall, receiver operating curve (ROC) area,
and F-­measure are also considered in this table. As our dataset contains no outliers,
we will use accuracy as our primary parameter for evaluating our classifier’s effec-
tiveness [47–57].
 Evaluation of machine learning algorithms  79

Table 4.2  Accuracy achieved by classification algorithm with all features of

dataset

Classification algorithm Accuracy (%) MAE Precision Recall

Naïve Bayes + k-­fold cross-­validation 86.59 0.063 0.870 0.866
Random forest + k-­fold cross-­validation 92.14 0.118 0.924 0.921
Decision tree + k-­fold cross-­validation 96.76 0.017 0.968 0.968
MLP + k-­fold cross-­validation 84.59 0.078 0.846 0.846
DT + k-­fold cross-­validation 90.13 0.146 0.909 0.901
JRip + k-­fold cross-­validation 96.14 0.024 0.962 0.961
LR + k-­fold cross-­validation 87.51 0.049 0.877 0.875

When it comes to classifying datasets, accuracy is defined as the total num-

ber of correctly predicted items split into the total number of predictions generated
by the algorithm. Graphical representation of k-­fold cross-­validation (classification
algorithms with 10 cross-­validation) implementation is shown in the figure below.
According to the graph, when compared to other algorithms, the decision tree clas-
sification algorithm performs remarkably well.
The accuracy of a classification algorithm is defined as the ratio of correct pre-
dictions to total predictions generated by the algorithm for a particular dataset. Thus,
accuracy as a performance metric for an algorithm is inadequate for issues involving
imbalanced categorization [58–65]. The primary difficulty with the imbalance clas-
sification problem is that the overwhelming number of examples from the majority
class will trump the minority class’s examples. Which means that even the most
unskilled models can get accuracy scores of up to 100%, depending on the severity
of the class imbalance.

4.4.2 Implementation after attribute selection using ranker method

In this section, we will use the ranker search method to find the best attribute from
the student’s performance dataset that has the greatest effect on the classification
algorithm’s prediction accuracy. We attempted three alternative attribute evaluator
methods with ranker search: CorrelationAttributeEval, InfoGainAttributeEval, and
GainRatioAttributeEval. Following the implementation of the attribute evaluator
algorithm, we picked ten distinct attributes that had the most influence on our pre-
diction outcome. Table 4.3 illustrates the outcome of the complete procedure when
applied to the student dataset. The first ten selected attributes from each technique
are used to evaluate students’ academic achievement and are determined to be the
most predictive.
To identify the best classifiers that more accurately generalize the data, we
apply various classification algorithms to a student performance dataset, selecting
the parameters of the algorithms that most effectively analyze the data and enhance
their generalization accuracy. Naïve Bayes, random forest, J48, MLP, DT, JRip,
and LR are all implemented classification methods. Please keep in mind that all
80  Intelligent network design driven by Big Data analytics

implemented classification algorithms undergo a 10-­fold cross-­validation to deter-

mine the algorithms accuracy when different parameter selections are used.
Implementation of classification algorithm after CAE FS: Classification,
grouping, and regression algorithms all utilize a training dataset to establish weight
factors that may be applied to previously unseen data for predictive purposes. Prior
to executing a data mining technique, it is required to narrow down the training
dataset to the most relevant attributes. Dimensionality reduction is the process of
modifying a dataset in order to extract only the characteristics required for training.
Due to its simplicity and computational efficiency, dimension reduction is critical
since it minimizes overfitting. Thus, dimensionality reduction is critical through-
out the data preprocessing phase. A correlation-­based FS method selects attributes
based on the usefulness of individual features for predicting the class label, as well
as the degree of connection between them. We avoid strongly linked and irrele-
vant features. The CAE determines an attribute’s value in a dataset by calculating
the correlation between the attribute and the class attribute. Nominal qualities are
assessed individually, with each value acting as a signal. A weighted average is used
to generate an overall correlation for a nominal characteristic. We picked the top ten
attributes with a threshold value larger than one using the aforementioned attribute
evaluator CAE in conjunction with the ranker search strategy.
The following table summarizes the results of the implementation of several
classification algorithms using CAE and the test option as k-­fold cross-­validation
approaches. As shown in Table 4.4, the combination (LR + CAE + k-­fold cross-­
validation) achieved the greatest accuracy of 97.84% when compared to other clas-
sification algorithms such as Naïve Bayes, random forest, decision tree, MLP, DT,
and JRip. As can be seen, the MLP technique improves accuracy to 97.68%, which
is significantly higher than the accuracy obtained without utilizing the FS approach.
The remainder of the algorithms are also accurate to an acceptable level. Only the top
15 attributes of the dataset are considered when implementing these methods. Other
performance metrics such as MAE, precision, and recall value are also considered in
this table. As our dataset contains no outliers, we will use accuracy as our primary
parameter for evaluating our classifier’s effectiveness.

Table 4.3   Attribute selection with the help of ranker search method

Attribute in decreasing order of their rank

Attribute evaluator method

CorrelationAttributeEval (CAE) 31,32,33,15,21,1,7,8,14,27,13,4,28,3,30,2,2

2,11,16,9,19,25,24,5,29,26,20,18,23,17,
12,10,6
InfoGainAttributeEval (IGAE) 32,33,31,15,21,1,9,7,14,11,28,27,3,10,4,16,
12,22,2,20,19,18,5,23,17,6,29,26,30,24,
8,25,13
GainRatioAttributeEval (GRAE) 32,33,31,15,21,1,3,7,14,16,28,27,4,9,11,12,
22,18,10,2,20,19,5,23,17,6,25,30,26,13,
24,29,8
 Evaluation of machine learning algorithms  81

Figure 4.2  Graphical representation of accuracy level of classification

algorithms with CAE

Below Figure 4.2, is a graphical illustration of the implementation of the clas-

sification algorithms discussed previously using CAE and cross-­validation (k-­fold
cross-­validation) as testing methods? The graph clearly demonstrates that the LR
algorithm outperforms all other algorithms considered. However, as illustrated in
Figure  4.1, practically all classification systems obtain a prediction accuracy of
greater than 90%.
Implementation of classification algorithm after IGAE FS: The strategy for
selecting terms with the highest information gain scores is called information gain
FS. Information gain quantifies the amount of information about a class prediction
that is accessible in bits if all that is available is the presence of a feature and its
associated class distribution. It determines the value of a characteristic by examining
its correlation to the class. Nominal qualities are analyzed value by value, with each
value acting as an indicator. A weighted average is used to determine the overall

Table 4.4   Accuracy achieved by classification algorithm with CAE

Accuracy
Classification algorithm (%) MAE Precision Recall

Naïve Bayes + CAE + k-­fold cross-­validation 87.51 0.058 0.878 0.875

Random forest + CAE + k-­fold cross-­validation 95.83 0.061 0.959 0.958
Decision tree + CAE + k-­fold cross-­validation 96.91 0.0169 0.969 0.969
MLP + CAE + k-­fold cross-­validation 97.68 0.0172 0.977 0.977
DT + CAE + k-­fold cross-­validation 90.13 0.146 0.909 0.901
JRip+ CAE + k-­fold cross-­validation 95.53 0.0252 0.956 0.955
LR + CAE + k-­fold cross-­validation 97.84 0.0087 0.979 0.978
82  Intelligent network design driven by Big Data analytics

Table 4.5   Accuracy achieved by classification algorithm with IGAE

Classification algorithm Accuracy (%) MAE Precision Recall

Naïve Bayes + IGAE + k-­fold cross-­validation 87.98 0.0569 0.884 0.880

Random forest + IGAE + k-­fold cross-­validation 94.60 0.0782 0.947 0.946
Decision tree + IGAE + k-­fold cross-­validation 96.91 0.017 0.969 0.969
MLP + IGAE + k-­fold cross-­validation 95.37 0.032 0.954 0.954
DT+ IGAE + k-­fold cross-­validation 90.13 0.146 0.909 0.901
JRip+ IGAE + k-­fold cross-­validation 95.99 0.0239 0.960 0.960
LR + IGAE + k-­fold cross-­validation 94.60 0.0232 0.946 0.946

correlation of a nominal property. It determines the value of an attribute by calculat-

ing the information gained relative to the class.
The following Table 4.5 summarizes the results of implementing various clas-
sification algorithms using the IGAE as the FS algorithm and k-­fold cross-­validation
as the test option method. According to Table 4.5, the (decision tree+ IGAE + k-­fold
cross-­validation) achieved the highest accuracy of 96.91% when compared to other
classification algorithms. As we can see, the MLP algorithm improves accuracy to
95.37%, which is significantly higher than the accuracy achieved without using the
FS method. The remainder of the algorithms are also accurate to an acceptable level.
Only the top 15 attributes of the dataset are considered when implementing these
methods.

Figure 4.3  Graphical representation of accuracy level of classification

algorithms with IGAE
 Evaluation of machine learning algorithms  83

Table 4.6   Accuracy achieved by classification algorithm with GRAE

Classification algorithm Accuracy (%) MAE Precision Recall

Naïve Bayes + GRAE + k-­fold cross-­validation 88.59 0.0565 0.889 0.886

Random forest + GRAE + k-­fold cross-­validation 93.83 0.074 0.939 0.938
Decision tree+ GRAE + k-­fold cross-­validation 96.91 0.017 0.969 0.969
MLP + GRAE + k-­fold cross-­validation 94.76 0.0283 0.948 0.948
DT+ GRAE + k-­fold cross-­validation 90.13 0.146 0.909 0.901
JRip+ GRAE + k-­fold cross-­validation 95.83 0.0232 0.959 0.958
LR + GRAE + k-­fold cross-­validation 95.83 0.017 0.959 0.958

Figure 4.3 illustrates the implementation of the above-­mentioned classification

algorithms using IGAE+ k-­fold cross-­validation. The graph demonstrates unequivo-
cally that the decision tree classification algorithm outperforms all other techniques
considered. However, as illustrated in Figure 4.1, practically all classification meth-
ods obtained greater than 87% prediction accuracy.
Implementation of classification algorithm after GRAE FS: The gain ratio
is a change in the information gain that has the effect of lowering bias. Gain ratio
is taken into account when selecting a characteristic since it takes into account the
number and size of branches. This method compensates for the information gain by
taking into account the intrinsic information contained within a split. The entropy
of a given instance’s distribution into branches is known as intrinsic information.
As the amount of intrinsic information rises, the value of an attribute decreases as
a result.
It determines the value of an attribute by calculating the gain ratio relative to the
class. The following table summarizes the outcomes of implementing several classi-
fication algorithms using the GRAE as the FS algorithm and k-­fold cross-­validation
as the test option method. According to Table 4.6, the decision tree+ GRAE + k-­fold
cross-­validation achieved the greatest accuracy of 96.91 %, when compared to other
classification algorithms. As can be seen, the MLP technique improves accuracy to
94.76%, which is significantly higher than the accuracy obtained without utilizing
the FS approach. The remainder of the algorithms are also accurate to an acceptable
level. Only the top 15 attributes of the dataset are considered when implementing
these methods.
Below Figure 4.4. is a graphical representation of the above-­mentioned clas-
sification algorithms implemented using the GRAE+ k-­fold cross-­validation
approach. The graph clearly demonstrates that the decision tree+ GRAE + k-­fold
cross-­validation technique outperforms all other algorithms considered. However,
as illustrated in Figure 4.1, practically all classification algorithms obtain a predic-
tion accuracy of greater than 88%.
84  Intelligent network design driven by Big Data analytics

Figure 4.4  Graphical representation of accuracy level of classification

algorithms with GRAE

4.5   Result analysis and discussion

A comparative examination of all of the implemented classification algorithms is

presented in Figure 4.5. To begin, we evaluate the prediction accuracy of classifica-
tion algorithms with FS (CAE, IGAF, and GRAF) approaches, using k-­fold cross-­
validation as a testing method.
As seen in the following table, we can analyze each method separately:
Naïve Bayes’ classification algorithm: k-­fold cross-­validation was used to test
three alternative FS strategies. Naïve Bayes + GRAE + k-­fold cross-­validation per-
formed incredibly well on the given dataset and achieved the maximum accuracy
of 88.59% in our observations. According to Table 4.7, the classification algorithm
that did not use FS method had the lowest accuracy of all FS methods (CAE, IGAE,
and GRAE). With the FS algorithm, the Naïve Bayes’ algorithm gains a maximum
improvement of more than 2%.
Random forest classification algorithm: Random forest + CAE + k-­fold
cross-­validation worked remarkably well on the supplied dataset, with a maximum
accuracy of 95.83%. According to Table 4.7, the classification algorithm without
employing FS method had the lowest accuracy of all FS methods (CAE, IGAE, and
GRAE).
Decision tree classification algorithm: Decision tree+ CAE + k-­fold cross-­
validation worked remarkably well on the supplied dataset, with a maximum
accuracy of 96.91%. According to Table 4.7, the classification algorithm without
employing FS method had the lowest accuracy of all FS methods (CAE, IGAE, and
GRAE).
 Evaluation of machine learning algorithms  85

Figure 4.5  Graphical representation of classification algorithm with accuracy

using CAE, IGAE, and GRAE

MLP classification algorithm: MLP + CAE + k-­fold cross-­validation per-

formed remarkably well compared to other combinations on the given dataset,
achieving maximum accuracy of up to 97.68% (see Table 4.7). According to the
table, the classification algorithm based on FS achieved the highest accuracy. The
FS approach provides a maximum improvement of more than 13% over MLP.

Table 4.7   Classification algorithm with accuracy using CAE, IGAE, and GRAE

Classification Accuracy Accuracy Accuracy

algorithms Accuracy (%) (CAE) (%) (IGAE) (%) (GRAE) (%)
Naïve Bayes 86.59 87.51 87.98 88.59
Random forest 92.14 95.83 94.60 93.83
Decision tree 96.76 96.91 96.91 96.91
MLP 84.59 97.68 95.37 94.76
DT 90.13 90.13 90.13 90.13
JRip 96.14 95.53 95.99 95.83
LR 87.51 97.84 94.60 95.83
86  Intelligent network design driven by Big Data analytics

DT classification algorithm From Table 4.7, it is clear that DT classification

algorithm without using FS method (CAE, IGAE, GRAE) achieved the same accu-
racy as with all features of the dataset.
JRip classification algorithm: When using a FS technique, the JRip algo-
rithm’s prediction accuracy is slightly reduced.
LR classification algorithm: LR + CAE + k-­fold cross-­validation performed
remarkably well compared to other combinations on the supplied dataset, achieving
the maximum accuracy of 97.84% (Table 4.7). According to the table, the classifi-
cation algorithm without employing FS method had the lowest accuracy of all FS
methods (CAE, IGAE, and GRAE). With the FS algorithm, the highest improve-
ment LR algorithm gains is more than 10%.

4.6   Conclusion

If we want to have a better knowledge of the algorithms, we need to conduct a thorough

evaluation of algorithms using multiple performance metrics. Performance criteria such
as accuracy, MAE, precision and recall, among others, are taken into account when devel-
oping the model. Different metrics are used to evaluate the predictability, goodness, and
error measurement of various algorithms before a final decision is made on a particular
method. Since each problem has its own unique collection of qualities and characteristics,
it is hard to anticipate in advance which performance metrics are better for certain chal-
lenges, according to the literature review. Since this is the case, multiple measurements
should be taken in order to improve the algorithm’s performance. Different data mining
techniques and FS procedures were evaluated on the dataset in question, with positive
results. Each FS strategy that is used to build a classification algorithm enhances overall
implementation performance, according to our findings.

References

[1] Wang X., Mei X., Huang Q., Han Z., Huang C. ‘Fine-­grained learning per-
formance prediction via adaptive sparse self-­attention networks’. Information
Sciences. 2021, vol. 545(2), pp. 223–40.
[2] Romero C., Ventura S. ‘Educational data mining and learning analytics: an
updated survey’. WIREs Data Mining and Knowledge Discovery. 2020, vol.
10(3), pp. 25–37.
[3] Iatrellis O., Savvas Ilias Κ.., Fitsilis P., Gerogiannis V.C. ‘A two-­phase ma-
chine learning approach for predicting student outcomes’. Education and
Information Technologies. 2021, vol. 26(1), pp. 69–88.
[4] Romero C., Ventura S. ‘Educational data mining: a review of the state of the
art’. IEEE Transactions on Systems, Man, and Cybernetics, Part C. 2010, vol.
40(6), pp. 601–18.
[5] Jauhari F., Supianto A.A. ‘Building student’s performance decision tree clas-
sifier using boosting algorithm’. Indonesian Journal of Electrical Engineering
and Computer Science. 2019, vol. 14(3), pp. 1298–304.
 Evaluation of machine learning algorithms  87

[6] Hamoud A.K., Hashim A.S., Awadh W.A. ‘Predicting student performance
in higher education institutions using decision tree analysis’. International
Journal of Interactive Multimedia and Artificial Intelligence. 2018, vol. 5(2),
pp. 26–31.
[7] Slater S., Joksimović S., Kovanovic V., Baker R.S., Gasevic D. ‘Tools for
educational data mining’. Journal of Educational and Behavioral Statistics.
2017, vol. 42(1), pp. 85–106.
[8] Sokkhey P., Okazaki T. ‘Hybrid machine learning algorithms for predict-
ing academic performance’. International Journal of Advanced Computer
Science and Applications. 2020, vol. 11(1), pp. 32–41.
[9] Saa A.A., Al-­Emran M., Shaalan K. ‘Mining student information system
records to predict students’ academic performance’. The International
Conference on Advanced Machine Learning Technologies and Applications
(AMLTA2019); Springer, Cham, Cairo, Egypt, 28–30 Mar; 2019. pp.
229–39.
[10] Yadav A., Alexander V., Mehta S. ‘Case-­based instruction in undergraduate
engineering: does student confidence predict learning’. The International
Journal of Engineering Education. 2019, vol. 35(1), pp. 25–34.
[11] Ruiz S., Urretavizcaya M., Rodríguez C., Fernández-­Castro I. ‘Predicting stu-
dents’ outcomes from emotional response in the classroom and attendance’.
Interactive Learning Environments. 2020, vol. 28(1), pp. 107–29.
[12] Raga R.C., Raga J.D. ‘Early prediction of student performance in blended
learning courses using deep neural networks’. 2019 International Symposium
on Educational Technology (ISET); Hradec Kralove, Czech Republic, 2–4
Jul; 2019. pp. 39–43.
[13] Kuzilek J., Vaclavek J., Zdrahal Z., Fuglik V. ‘Analysing student VLE be-
haviour intensity and performance’. European Conference on Technology
Enhanced Learning; Springer, Cham, Delft, The Netherlands, 16–19 Sep;
2019. pp. 587–90.
[14] Sokkhey P., Okazaki T. ‘Comparative study of prediction models on high
school student performance in mathematics’. 34th International Technical
Conference on Circuits/Systems, Computers and Communications (ITC-­
CSCC); IEEE, JeJu, Korea (South), 23–26 Jun; 2019. pp. 1–4.
[15] Sharma A., Yadav D.P., Garg H., Kumar M., Sharma B., Koundal D. ‘Bone
cancer detection using feature extraction based machine learning model’.
Computational and Mathematical Methods in Medicine. 2021, vol. 2021, pp.
1–13.
[16] Kumar M., Bajaj K., Sharma B., Narang S. ‘A comparative performance as-
sessment of optimized multilevel ensemble learning model with existing clas-
sifier models’. Big Data. 2021
[17] Moubayed A., Injadat M., Shami A., Lutfiyya H. ‘Student engagement level
in an e-­learning environment: clustering using k-­means’. American Journal
of Distance Education. 2020, vol. 34(2), pp. 137–56.
[18] Moubayed A., Injadat M., Shami A., Lutfiyya H. ‘Relationship between stu-
dent engagement and performance in e-­learning environment using association
88  Intelligent network design driven by Big Data analytics

rules’. IEEE World Engineering Education Conference (EDUNINE); Buenos

Aires, Argentina, 11–14 Mar; 2018.
[19] Yadav S.K., Pal S. ‘Data mining: a prediction for performance improvement
of engineering students using classification’. Journal WCSIT. 2012, vol. 2(2),
pp. 51–6.
[20] Ramesh V., Parkavi P., Ramar K. ‘Predicting student performance: a statistical
and data mining approach’. International Journal of Computer Applications.
2013, vol. 63(8), pp. 35–9.
[21] Hung J.-L., Shelton B.E., Yang J., Du X. ‘Improving predictive modeling for
at-­risk student identification: a multistage approach’. IEEE Transactions on
Learning Technologies. 2019, vol. 12(2), pp. 148–57.
[22] Helal S., Li J., Liu L., et al. ‘Predicting academic performance by consider-
ing student heterogeneity’. Knowledge-­Based Systems. 2018, vol. 161, pp.
134–46.
[23] Xu J., Moon K.H., van der Schaar M. ‘A machine learning approach for track-
ing and predicting student performance in degree programs’. IEEE Journal of
Selected Topics in Signal Processing. 2017, vol. 11(5), pp. 742–53.
[24] Sekeroglu B., Dimililer K., Tuncal K. ‘Student performance prediction and
classification using machine learning algorithms’. 2019 8th International
Conference on Educational and Information Technology, ICEIT 2019,
Association for Computing Machinery; Cambridge, 2–4 Mar; 2019. pp. 7–11.
[25] Prabha S.L., Mohamed Shanavas A.R. ‘Educational data mining applica-
tions’. Operations Research and Applications: An International Journal.
2014, vol. 1(1), pp. 23–9.
[26] Baker R., Yacef K. ‘The state of educational data mining in 2009: a review
and future visions’. Journal of Educational Data Mining. 2009, vol. 1(1), pp.
3–17.
[27] Romero C., Ventura S. ‘Educational data mining: a review of the state of the
art’. IEEE Transactions on Systems, Man, and Cybernetics, Part C. 2010, vol.
40(6), pp. 601–18.
[28] Silva C., Fonseca J. ‘Educational data mining: a literature review’. Advances
in Intelligent Systems and Computing. 2016, vol. 520, pp. 87–94.
[29] Ahmed A.B.E.D., Elaraby I.S. ‘Data mining: a prediction for student’s perfor-
mance using classification method’. World Journal of Computer Application
and Technology. 2014, vol. 2(2), pp. 43–7.
[30] Baradwaj B.K., Pal S. ‘Mining educational data to analyze student’s per-
formance’. International Journal of Advanced Computer Science and
Applications. 2011, vol. 2(6), pp. 63–9.
[31] Archana S., Elangovan K.D. ‘Survey of classification techniques in data min-
ing’. International Journal of Computer Science and Mobile Applications.
2009, vol. 2(2), pp. 65–71.
[32] Khan I., Al Sadiri A., Ahmad A.R., Jabeur N. ‘Tracking student performance
in introductory programming by means of machine learning’. 4th MEC
International Conference on Big Data and Smart City (ICBDSC); Muscat,
Oman, 15–16 Jan; 2019. pp. 1–6.
 Evaluation of machine learning algorithms  89

[33] Al-­Razgan M., Al-­Khalifa A.S., Al-­Khalifa H.S. ‘Educational data mining:
a systematic review of the published literature 2006-­2013’. Proceedings
of the First International Conference on Advanced Data and Information
Engineering (DaEng-­2013); Singapore: Springer; 2014. pp. 711–19.
[34] Bhardwaj B.K., Pal S. ‘Data mining: a prediction for performance improve-
ment using classification’. Computer Science Information Retrieval. 2012,
vol. 9(4), pp. 136–40.
[35] Baker R.S., Yacef K. ‘The state of educational data mining in 2009: a review
and future visions’. Journal of Educational Data Mining. 2009, vol. 1(1), pp.
3–17.
[36] Romero C., Ventura S. ‘Educational data mining: a review of the state of the
art’. IEEE Transactions on Systems, Man, and Cybernetics, Part C. 2010, vol.
40(6), pp. 601–18.
[37] Silva C., Fonseca J. ‘Educational data mining: a literature review’. Europe
and MENA Cooperation Advances in Information and Communication
Technologies. Cham: Springer; 2017. pp. 87–94.
[38] Tsai C.-F., Tsai C.-T., Hung C.-S., Hwang P.-S. ‘Data mining techniques for
identifying students at risk of failing a computer proficiency test required
for graduation’. Australasian Journal of Educational Technology. 2011, vol.
27(3), pp. 481–98.
[39] Li J., Li P., Niu W. ‘Artificial intelligence applications in upper gastrointesti-
nal cancers’. The Lancet. Oncology. 2020, vol. 21(1).
[40] Ma W., Adesope O.O., Nesbit J.C., Liu Q. ‘Intelligent tutoring systems and
learning outcomes: a meta-­analysis’. Journal of Educational Psychology.
2014, vol. 106(4), pp. 901–18.
[41] Macfadyen L.P., Dawson S., Pardo A., Gasevic D. ‘Embracing big data in
complex educational systems: the learning analytics imperative and the
policy challenge’. Research & Practice in Assessment. 2014, vol. 9, pp.
17–28.
[42] Mussack D., Flemming R., Schrater P., Cardoso-­Leite P. ‘Towards discover-
ing problem similarity through deep learning: combining problem features
and user behaviour’. Proceedings of the 12th International Conference on
Educational Data Mining (EDM 2019); Educational Data Mining Forum,
2019. pp. 615–18.
[43] Pu Y., Wu W., Jiang T. ‘ATC framework: A fully automatic cognitive tracing
model for student and educational contents’. Proceedings of 12th International
Conference on Educational Data Mining; Educational Data Mining, 2019.
pp. 635–38.
[44] Lahoura V., Singh H., Aggarwal A., et al. ‘Cloud computing-­based framework
for breast cancer diagnosis using extreme learning machine’. Diagnostics.
2021, vol. 11(2), p. 241.
[45] Koundal D., Sharma B. 'Challenges and future directions in neutrosophic set-­
based medical image analysis' in Neutrosophic set in medical image analysis.
Academic Press; 2019. pp. 313–43.
90  Intelligent network design driven by Big Data analytics

[46] Singh K., Sharma B., Singh J., et al. ‘Local statistics-­based speckle reduc-
ing bilateral filter for medical ultrasound images’. Mobile Networks and
Applications. 2020, vol. 25(6), pp. 2367–89.
[47] Lahoura V., Singh H., Aggarwal A., et al. ‘Cloud computing-­based framework
for breast cancer diagnosis using extreme learning machine’. Diagnostics.
2021, vol. 11(2), p. 241.
[48] Haidar M., Kumar S. ‘Smart healthcare system for biomedical and health care
applications using aadhaar and blockchain’. 5th International Conference on
Information Systems and Computer Networks, ISCON 2021; Mathura, IEEE,
2022. pp. 1–5.
[49] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
Personal Communications. 2021, vol. 10(3), pp. 1–14.
[50] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[51] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[52] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering (IJITEE). 2019, vol. 8(6), pp.
1672–6.
[53] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based Sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[54] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[55] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring
using machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[56] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and iot for
smart road traffic management system’. IEEE India Council International
Subsections Conference (INDISCON); Visakhapatnam, India, IEEE, 2020.
pp. 289–96.
[57] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation
of fault tolerance technique for internet of things (iot)’. 12th International
Conference on Computational Intelligence and Communication Networks
(CICN); Bhimtal, India, IEEE, 2020. pp. 154–59.
 Evaluation of machine learning algorithms  91

[58] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. 10th International
Conference on Cloud Computing, Data Science & Engineering (Confluence);
Noida, India, IEEE, 2020. pp. 63–76.
[59] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 4th International Conference on Information Systems and
Computer Networks (ISCON); Mathura, India, IEEE, 2019. pp. 250–55.
[60] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
International Conference on Computational Intelligence and Communication
Networks (CICN); Jabalpur, India, IEEE, 2016. pp. 79–84.
[61] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress
in Electromagnetics Research Symposium; Prague, Czech Republic, The
Electromagnetics Academy, 2015. pp. 2363–67.
[62] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wireless
sensor network’. Fifth International Conference on Communication Systems
and Network Technologies; Gwalior, India, IEEE, 2015. pp. 194–200.
[63] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distributed lo-
calized wireless sensor networks’. 2014 International Conference on Issues
and Challenges in Intelligent Computing Techniques (ICICT), Publisher:
IEEE; Ghaziabad, India, 7–8 Feb; 2014.
[64] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­ based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms
for Intelligent Systems; THDC-­ Institute of Hydropower Engineering and
Technology, Tehri, India, Singapore: Springer, 2020. pp. 327–36.
[65] Kumar S. Evolution of software-­defined networking foundations for iot and
5G mobile networks. Hershey, PA: IGI Publisher; 2020. p. 350.
This page intentionally left blank
 Chapter 5
Accurate management and progression of Big
Data Analysis
Tanmay Sinha 1, Narsepalli Pradyumna 1, and K B Sowmya 1

Statistical investigations are concerned with circumstances including arranging,

information assortment, association of data, and scrutiny of compiled data, conver-
sion, and exposure in a clear and chosen procedure. To do so, research techniques
can be distinguished in two different ways: conclusion overviews and statistical sur-
veying. In assessments of public sentiment, the primary objective is to assemble data
about deciding subjects dependent on close-­to-­home meetings. Statistical surveying
is directed through the market investigation of a specific item. Descriptive statistics
are responsible for collection, association, depiction of information, estimation, and
translation of coefficients, whereas inductive or inferential statistics, also known as
the proportion of vulnerability or techniques that rely on likelihood hypothesis, are
responsible for investigation and understanding of information related to an edge
of vulnerability. In statistics, we look at how to use tables and charts. The tables
organise and classify the data, while the graphics convey the information in a clear
and easy manner, aiding in goal attainment.

5.1 Introduction

In today’s world, most things are digital in nature. Every nine out of ten works are
done digitally from buying vegetables to booking shows and so on. Our every digital
transaction needs to be stored for safety reasons and for customer satisfaction and
their preferences. So, a digital era indicates an enormous amount of data to be stored
and processed. Therefore, we can also say that this is an era of data.
First of all, we will start with the definition of data which is a combination
of the amount, pictures, or characters on which tasks are executed by a personal
computer that can be renounced and passed on in the form of electrical signals and
subsequently taped on attractive, optical, or mechanical account media. Big Data is
also a kind of data having immense dimensions. Big Data is a term, which depicts

1
Department of Electronics and Communication Engineering, Bengaluru, India
94  Intelligent network design driven by Big Data analytics

an array of information having enormous volume but increasing exponentially with

increasing time. In a nutshell, this information is so enormous along with the com-
plexity that no other traditional data the executive’s device can cache or process
in a fruitful manner. Let it be whether to increase customer friendliness or ease of
deduction by the officials, we need to process and store data. This process is called
Big Data Analysis.
In every place, whether service-­oriented or not, data need to be stored. There are
several important fields such as e-­commerce, finance, education, healthcare, bank-
ing, social networking, purchase at department/grocery stores, and so on. Day-­by-­
day, the exposure to such tremendous data is becoming increasingly common and
one can achieve insights for supporting high-­quality business choices and can form
more experienced and customised higher quality services. For gathering data as per
the requirement, MapReduce program is used enormously. To operate such Big Data
and to attain greater performance, proper scheduling is essential. This scheduling is
basically a job-­giving technique to available resources in a way so that starvation
can be reduced to a large extent and resources that are abundant can be utilised to a
maximum. Enormous information is a term generally used for the processes that are
unconventional and discoveries that are assumed to accumulate, compose, process,
and arrange experiences from data sets that are extremely big. Presently, there will
be discussion and gathering of extensive information at the crucial level that will
specify the normal ideas which one may spill over in mind during investigation of
the subject. Furthermore, we will ponder upon the procedures and advancements,
which are employed right now in this process.
For the storage of huge quantity of information to reveal hidden patterns, cor-
relations, and other insights, Big Data is used extensively. At present, analysis of
data and decoding with immediate responses is possible, which is an attempt that is
slower in rate and has less efficiency with more conventional business intelligence
solutions. Big Data Analytics help organisations utilise their information and new
opportunities are identified with usage of the same. A huge amount of information
in a database can be compiled using Big Data, and retrieval of data and information
from that database is also easy.

5.1.1  Examples of Big Data

As we have discussed, in the present data era, we have enormous data from various
sources that need to be processed for proper usage and storage. Some of the exam-
ples of Big Data are as follows:

• Social Media

1. According to a measurement, social media can store more than 500 terabytes
of information, which are new and cached into the databases of Facebook. This
information is later created in the form of photographs along with video trans-
fers, message trades, and also enables putting remarks, and so on.
 Accurate management and progression of Big Data Analysis  95

2. Enables sending 500 million tweets

3. Enables sending 294 billion emails
4. Creation of 4 petabytes of data on Facebook
5. Creation of 4 terabytes of data using each connected car
6. Enables sending 65 billion messages on WhatsApp

• Enable almost 5 billion searches

• Stock markets in India generate about 1 terabyte of data every day.
• All our daily usage apps/websites have Big Data as we can easily say that com-
panies like Amazon and Uber and banking apps and websites, etc., have more
than a million customers; data regarding each and every product they are selling
along with their sellers and data regarding various functionalities like shipment
tracking, etc.
• We can consider various other applications like a self-­driving car, UAVs, moni-
toring systems, anomaly detection, and many others, as depicted in Figure 5.1.

Figure 5.1   Big Data application areas

96  Intelligent network design driven by Big Data analytics

5.2 Big Data Analysis

Big Data is never accessed without proper processing since it makes the usage of
data very difficult. Processing steps are used for identifying hidden patterns, market
trends, and consumer preferences for the benefit of organisational decision-­making.
The first step for Big Data Analysis is collecting the data but the raw data we
collect from our sources may it be some websites or apps or any other platforms are
unusable. We need to process and convert the enormous data set into meaningful
data, which can be used by the organisation. This step is called Data Mining. This
step is also called Knowledge Discovery in Databases. In this step, we derive crude
but essential information from the data set. There are various methods for doing this
like k-­nearest neighbors (KNN), clustering, regression, association rules, sequential
patterns, etc.
The second step is the actual analysis of processed data depending upon the
need. We can use various machine learning algorithms and select the one/more data
columns for solving the problem.
This combination of steps is called Big Data Analytics, i.e., Data collection –
Raw data, Data mining – Mined data, and Data analysis.
Mined data can be classified as follows (see Figure 5.2):

• statistics
• machine learning
• visualisation
• information science
• database technology
• other disciplines

5.2.1  Life cycle of Big Data

The history of Big Data lies around the late nineteenth century. The first data overload
came into picture in 1881 while calculating the census of 1880. To solve all the prob-
lems faced, Hollerith Tabulating Machine was developed which cut the processing

Figure 5.2   Uses of mined data

 Accurate management and progression of Big Data Analysis  97

Figure 5.3   Different stages of Big Data Analysis

of data from 10 years labour work to 1 year. Following that, a German-­Austrian

Engineer Fritz Pfleumer developed a magnetic data storage on tape in 1928. This
magnetic data storage was believed to store digital data. Another notable moment
in the history of Big Data came into picture in 1948 where Shannon’s information
theory was evolved which laid the very basic functional block of today’s informa-
tion infrastructure. In 1970, Edgar F. Codd from IBM gave ‘relational database’
which explained how the information in the large databases can be accessed without
identifying the source or the structural framework. In 2001, the main characteris-
tics of Big Data, i.e., three Vs were presented in the paper by Doug Laney thereby
paving the basic framing of Big Data further ahead. In the same year, software-­as-­
a-­service was also shared with the public. After developing these many foundation
blocks in the direction of Big Data, there was a need for an open source software
utility, which would act as a framework for such data. So, Hadoop was developed in
2005. Before 2007, no name was given to such humongous data. In a Wired Article:
‘The End of Theory: The Data Deluge Makes the Scientific Method Obsolete’, this
humongous data were given a name, i.e., Big Data. The stage so obtained in 2007
was the beginning of the era of Big Data Computing. In 2017, IBM study analysed
that around 2.5 quintillion of data are created every day and around 90% of the total
world’s data is created in just 2 years. This is one of the most remarkable works done
in the field of Big Data.
From different sources such as web, sensors, Internet of things (IoT), social
media, mobile data, etc., Big Data is collected and then it is stored in different data-
base systems such as RDBMS, File Storage System, etc. [1–10].
After the storage process, different processing techniques are applied and the
large data sets stored are processed; processing techniques include MapReduce,
Mahout, etc. After processing, analysis of data is done and finally it is archived with
Cloud Backup or Amazon Glacier.
This whole process is shown in Figure 5.3.

5.2.2  Classification of the Big Data

Broadly, classification of the Big Data can be done in three major domains, as shown
in Figure 5.4.

• Structured
98  Intelligent network design driven by Big Data analytics

Figure 5.4   Classification for a Big Data Analysis

Organised information can be coined as a group of information, which can be

set aside, directed, and handled as a properly fixed arrangement. With the passage
of time, the software engineering abilities have progressed to a greater height and
has led to processed formation which enables working with such required pieces
of information (in this, configuration is more noticeable than time) and later on
enables drawing a valid reason out of it. Besides, several issues have been developed
due to procurement of an enormously large size of information having immense
degree, which has led to shortening of common places due the fierceness of different
zettabytes, which has led to large sizes of data [11–17].

• Unstructured

Basically, if any data whose form or structure is quite problematic to under-

stand, such data or information can be termed as unstructured statistics. Despite the
size being greater than any other data, there are quite a lot of problems or issues rep-
resented from this unstructured data as a way of preparing for figuring out a proper
result out of the same. Coming to case studies, one of the most typical instances of
unstructured data on record is assumed to be heterogeneous data delivery, which
involves a combination of confidential and important content records, collection
of snap shots and recordings, and few more data in the similar fashion. Nowadays,
there is high abundance of data, which is readily available at the same moment with
the institutions or the owners, or it can be said that the required basic data are avail-
able with the owners at any moment of time but the only difficulty is that they do not
have the rough idea to get the required result or an incentive from that collective data
keeping into consideration that the statistics available is in original or crude form or
it can be said an unstructured organisation of the data [18–20].

• Semi-­structured

Coming to semi-­structured data, it includes both forms of statistics. So semi-­

structured can be designed as an organised or structured data structure but it cannot
be said that it is structured to a great extent. Taking an instance of desk definition
in DBMS, it can be said that semi-­structure-­based data is not at all structured data.
 Accurate management and progression of Big Data Analysis  99

Basically, semi-­structured data can be defined or said to be information, which is

transcripted in an XML document.

5.2.3  Working of Big Data Analysis

Here in Figure 5.5, a typical flowchart has been given on how Big Data Analysis
performs its job. It includes collection, processing, cleaning, and analysis of huge
data sets which may help organisations to use their Big Data. We will see each pro-
cess one by one.

a. Collection of data: In this, data from various sources are collected. Data may
be structured or unstructured and may vary from cloud to mobile applications.

Figure 5.5   Typical flowchart for Big Data Analysis

100  Intelligent network design driven by Big Data analytics

These data will be stored in data warehouses from where it can be accessed with
greater ease. Unstructured data may be stored in a data lake.
b. Processing of data: After the data are stored, the next step is to undergo analysis
and then organise such that accuracy can be maintained. For processing the
data, there are two ways which can be followed while doing, viz., batch process-
ing and stream processing.
c. Cleaning of data: After processing of data, it is needed to enhance the quality of
data and get more accurate results. If this cleaning of data is done properly, any
duplicated data or irrelevant data will be eliminated.
d. Analysis of data: After cleaning of irrelevant data, we are left with the actual
data needed. But still, conversion of those data sets into usable data might take
time. For such purposes, we have few data analysis tools which may be fol-
lowed to analyse the data more efficiently. Some of the data analysis tools are
listed below:

i) Data mining
ii) Predictive analysis
ii) Deep learning
Implementation of Big Data Analytics can be understood from Figure 5.6.

5.2.4  Common flaws that undermine Big Data Analysis

• The size of the sample may not reflect the target population or process.
• All the data collected with time may not reflect the current scenario due to which
efficiency of applying analytics to the situation reduces. There are many prob-
lems for proper management and progression of Big Data Analysis [21–24].

5.2.5  Advantages of Big Data Analysis

• Creative ideas and solutions can be derived from Big Data. Such creativity
leads to better understanding of what the customer wants and targeting them
more efficiently. Such data may be helpful for business purposes. Although it
has the advantage of storing huge amounts of data within it and targeting cus-
tomers more efficiently, there is one more advantage of such Big Data is that it
helps in creating more growth opportunities. This may also lead to development
of few new business areas [25–30].
• In the case of science and research, Big Data has led to great developments.
Data science, one of the most trending topics nowadays, is possible only due to
breakthroughs in the Big Data field since the more the data, the more accurate
the prediction with proper algorithms.
• Real-­time predictions, monitoring, and flaw detection are possible due to Big
Data Analysis.
• Due to the nature of Machine Learning and Artificial Intelligence, Big Data
Analytics is becoming an essential entity in the field.
 Accurate management and progression of Big Data Analysis  101

Figure 5.6   Big Data Analysis implementation layer

5.3. Processing techniques

In processing Big Data, three major characteristics are taken into consideration.
Those characteristics are volume of the data, velocity of the processing of the data,
and wide variety of the data.

5.3.1  Traditional method

An example for traditional parallel and distributed processing involves huge data,
e.g., top Spotify songs from 2010 to 2019 – by year. This data set can be found in​
Kaggle.​com; in this data set, let us take a task like identifying the most liked music
genre [31–34].
For completing this task using the traditional method:
102  Intelligent network design driven by Big Data analytics

1. Step 1: The whole data are split into small blocks and stored in different systems.
2. Step 2: Find the highest recurring genre for each part stored on the correspond-
ing machine.
3. Step 3: Results are taken from each machine to produce the final output.

Some of the problems discovered in traditional methods are as follows:

1. Critical path problem: The path may get extremely long for some applications
due to dividing data without proper management. The time which it takes to fin-
ish the task without delaying the next milestone or the actual completion date.
Therefore, if there are any delays, the entire work will be delayed.
2. Reliability problem: In case of any hardware or software problem, the data
will not be available which reduces its reliability. The management of this type
of problem is a challenge.
3. Splitting data issue: The way to divide the data among systems and the place
where to and how is something a problem; we need to find an efficient way to
split the data to have no burden on any device.
4. Single system may fail: There should be a way for fault tolerance. For example,
if one system is unable to produce the output for the final calculation, it should
still be able to produce results depending on the other outputs.
5. Accumulation of result: We need a mechanism to accumulate the results pro-
duced by each machine to produce the final output.

Due to the above problems, a programmer needs to take the design issues into
consideration while coding the application. To counter the problems that are present
in traditional methods, we have generated an innovative way, i.e., MapReduce. It
helps to take care of individually when using traditional approaches while process-
ing large amounts of data in parallel.
MapReduce framework can be used by programmers to create complex applica-
tions with parallel computations without taking system issues into consideration.

5.3.2 MapReduce
MapReduce is a programming environment that allows for distributed and parallel
processing of large data sets in a distributed environment.
As in its name, this framework is bi-­staged:

1. Mapping
2. Reducing

It follows the same order as in its name: The first being comparison and the next
being editor phase.
So the first is the task of comparing and creating key-­value pairs based on count.
This is the intermediate output of the process. This output is then entered into the
editor.
 Accurate management and progression of Big Data Analysis  103

The editor gets a pair of key values from several map jobs. The adjuster then
consolidates these intermediate data sets (the intermediate key-­value pair) into a
smaller set of motors or key-­value pairs, which is the final result.
As just mentioning the way cannot explain the process, let us take an example.
Let us take one of the most popular examples: word count.
In this type, we will take a text file with many words in it. Let us assume that the
content of the text file be as below:

Car, Bike, Cycle, Cycle, Bike, Car, Bike, Car, Bike, Cycle
The whole process can be divided into five steps

1. Splitting the data

2. Mapping the data
3. Shuffling the data
4. Reducing the data
5. Final result

Step 1: Splitting
Breaking the input into many chunks of data is done in this step. In this example,
let us take the number of chunks to be three for ease of understanding.
Step 2: Mapping
Tokenise each mapper word and assign each token or word a hard-­coded value
(1). The reason for setting the hard-­coded value to one is because we consider each
word to be unique and appear once. This creates a list of key-­value pairs, and the
key is a single unit. Words and values are one. Therefore, in the first row (Car, Bike,
Cycle), there are three key-­value pairs – Car(1), Bike(1), Cycle(1). This process is
repeated at all the nodes.
Step 3: Shuffling
The sorting and shuffling partition process is executed and all the same keyed
ordered pairs are sent to each corresponding reducer. After this phase, each reducer
has a list of values and a unique key corresponding to each value.
Step 4: Reducing
Now the reducer counts the values that exist in its list of values. As shown in the
figure, reducer gets a list of values that are [1, 1, 1] for key Car. It then counts the
number of 1 second in the list and returns the final output as – Car, 3.
Step 5: Final output
All output pairs are then collected and written to the output file. The process is
pictorially shown in Figure 5.7.

5.3.3  Advantages of MapReduce

1. Parallel processing
2. Data locality
104  Intelligent network design driven by Big Data analytics

Figure 5.7   Word count overview

Parallel processing
MapReduce divides the job into multiple nodes, each node processing part of the job
at the same time. As such, MapReduce is based on the divide-­and-­conquer paradigm
and is useful for processing data at different nodes. Processing data in parallel at
multiple nodes instead at a single node as conventional methods suggest signifi-
cantly reduces the time it takes to process the data. It can also be called distributed
processing as processing is distributed among different systems.

Data locality
In the case of data locality, the transfer of data from distinct data nodes to a single
node, which is the processing unit for various processes does not happen at all. The
nodes which are the processing units are further shared with the data nodes so as to
make sure that the processing is going to take place at data nodes only when the case
of MapReduce is taken into account.
Assuming the case of traditional systems, the data are first brought near to a
processing device to process the same. But if the data become too large to be pro-
cessed, sharing the same to processing nodes becomes problematic and thus causes
problems, which are listed below:

• Poor cost-­to-­network performance ratio. It is costly to share data between nodes

and it degrades the network performance.
• Processing the data at a single node causes a lot of delay.
• Processing nodes may overload and fail.

At present, MapReduce is one of the solutions, which is utilised to tackle the

problem stated above by involving or deploying the units of processing into the data
considered.
Thus, it can be observed in Figure 5.8 that the considered data are further shared
among numerous nodes wherein each node is performing the processing step, which
means processing some amount of data there itself.
Coming to the benefits associated with the same, few of them are listed below:
 Accurate management and progression of Big Data Analysis  105

Figure 5.8   MapReduce approach for processing and data locality

1. Moving the processor to the data is very cost effective.

2. Reduced processing time.
3. Removes overburdening of any node.

5.4 Cyber crime

The first thing to be understood is cybercrime. Cybercrime can be explained as a

virtual weapon to do illegal work relating to online stuff. This online stuff includes
committing fraudism and child trafficking in unseen videos. In the present world,
cybercrime has grown quite rapidly, which has a significant impact on the gross
business and the economy worldwide. Digital hoodlums who are using the complex
assault are growing each day in a progressive manner, and the hostile techniques uti-
lised by them have become inadequate with respect to the development that is going
on in the fields of knowledge and insights associated with it. Taking Big Data into
account, it is doing a huge transformation in the perspective, which leans towards
the developments that has been done or is going on in dangers relating to cyber-
crime. Coming to private or confidential data, each individual person possesses a
unique identification card. This card encloses each and every detail of that individual
starting from name and date of birth to the fingerprints. As a result, all the informa-
tion of an individual can be accessed by anyone. This information can be accessed
by the expert hackers and the information might be hacked. The mechanical discov-
ery that has been done in the field of Big Data has made it feasible and accountable
to handle huge data that are expanded and unstructured. Those kinds of data include
secured military operation, stockpilation, accumulation, and handling the charac-
terised scope in a progressive manner. This examination of Big Data empowers the
privileged arrangements done for the information and the knowledge of the risk to
be known, keeping aside the malevolent or the suspicious exercises performed by
the potentially distorted or the monstrous informational indices. Current advance-
ments in the direction of Big Data flash the mechanical control for concise reactions
106  Intelligent network design driven by Big Data analytics

additionally to acknowledge or justify the cybercrime dangers. One of the examples

is upsetting obviously distinguished malware assaults.
There are many branches which deal with this problem of cybercrime. Few of
them are listed here:

• Usage of Big Data in forensics

• Taking the control over phishing and pharming using Big Data Analytics
• Usage of the architectures of the Big Data for security purposes especially
cybersecurity
• Accounting Big Data with respect to criminal usage of IoT
• Cryptography and humongous data
• Privacy preserving retrieval, transmission, processing, and analysis of huge data
• Data-­intensive detection and prevention of online identity theft
• Access control of huge data
• Big Data Analytics for online social network threats
• Analysis of the malware and botnets by using Big Data
• Involving the techniques of the Big Data in the detection of the intrusion
• Using data mining and machine learning in anti-­cybercrime
• Development of formal models and ontologies for taking into account online
criminal behaviour
• Criminal abuse of huge data technologies and resources
• Steganography analysis and covert/subliminal channels
• Usage of Big Data in Denial of service (DoS)/distributed denial of service
(DDoS) Défense
• Adversary-­resilient Big Data technologies
• Standardisation advances in Big Data for cybersecurity

5.4.1  D
 ifferent strategies in Big Data to help in various
circumstances
Due to the rapid development in data science, it is difficult to sort apparatus and
strategies systematically. Few of them are reviewed here.
MapReduce Basically, this can be stated as a design which is used for huge
tasks parallely that can be performed efficiently. The efficiency of such a structure
depends on the deployment of the internal correlation in the data sets considered.
Normally, such a case is not accounted for when digital evidence is the central point
involved, rather a part of the file classified which looks similar to a task suits effi-
ciently when portrayed in the programming environment of MapReduce. When they
take different tasks into consideration, one of the most common among them in the
field of forensics is swapping of the parts of the file that is coming from the whole
file system picture or from unallocated space to file types, which are predefined.
These definite file types include algorithms involved in machine learning classi-
fication such as logistic regression, k-­means clustering, support vector machines
(SVM), and so on. These algorithms can be attached to the MapReduce paradigm
 Accurate management and progression of Big Data Analysis  107

if the individual involved in the analysis of the document tries to mold all the pos-
sibilities of the correlations among individual parts. A more consolidated approach
in the same direction can be merging the classification algorithms with a decision
tree. Such integration approach yields a higher accuracy with respect to individual
approach.
Decision trees and random forests These methods are generally brought to
give productive results in the case of extortion detection software. Here, the main
target is to identify the factual expectations or the predictions in the huge data set.
For such cases, situation odd exchanges or irregular pursuing conduct can be used
depending on the application and the usage.
Audio forensics Independent learning strategies under the overall meaning
of ‘blind signal separation’ give agreeable results in separating two superimposed
speakers or a sound from background noise in sound forensics. Among likely plans,
they rely upon numerical support to track down the base-­connected signs.
Image forensics Grouping techniques are helpful in picture legal sciences addi-
tionally to audit enormous arrangements of hundreds or thousands of picture records
as in to recognise dubious pictures from the rest.
Neural networks For complex example acknowledgment in network legal sci-
ences, neural networks are appropriate. This is fundamentally a helpful approach
where the progressive depictions of the document framework are utilised to prepare
the network to distinguish typical conduct of an application. Following the occasion,
the framework can be used subsequently to outline an execution course of events on
a legitimate picture of a record framework. Such neural networks are additionally
used to recognise network traffic yet do not yield significant degrees of exactness.
Another strategy is utilised, e.g., natural language processing procedures includ-
ing Bayesian classifiers and independent calculations for gathering like k-­means,
which is effective for origin confirmation or characterisation of huge groups of
unstructured texts, messages in explicit.

5.4.2 Big Data Analytics and cybercrime

Big Data Analytics is really helpful in tackling the issues related to cybercrime as
it is mainly developed for holding large volumes of data. It is much more than just
handling the data organisations and the reason why it is like that is because of the
well-­organised data algorithms that form data analysis structures. Also it is capa-
ble of identifying the variety of cybercrime either it is hacking or an online fraud
and this is done with the help of pattern recognition technology which identifies
the similarities by studying the data attack surroundings. Besides just tackling the
issue, Big Data Analytics has such a nice framework that it can predict the crime
before it occurs as it has the capability of reading data and drawing conclusions
from it. But this case it works well if artificial intelligence and machine learning
are incorporated into the same platform. This ability to alarm the organisations
before the attack of crime is boon for it as it may help in protecting the confidential
data and develop much better network guards. Stepwise representation is shown in
Figure 5.9.
108  Intelligent network design driven by Big Data analytics

Figure 5.9   Big Data Analytics in cybersecurity

5.4.3 Security issues associated with Big Data

As we all know that Big Data is visualised in four dimensions, i.e., Volume, Variety,
Velocity, and Veracity. Different mining algorithms and high-­performance com-
puting devices come into picture, and with the usage of such devices and algo-
rithms, analysis can be performed with negligible time or consumption of least time.
Keeping in mind all the advantages associated with Big Data, there are some high-­
level security issues relating to Big Data which is mentioned below:

a. real-­time analysis of data,

b. confidential mannered storage of transaction logs and data linked with it,
c. control access linked at granular level, and
d. data provenance.

For the analysis of humongous data, it is not compulsory to evaluate the same at
a very high speed ( approximately 30 00 00 000 m/s) rather it is important to store
data securely. These security methodologies involve encryption and decryption of
the data and these measures should be an integral part of the analysis because there
are quite a lot of unethical experts who can extract the data without notifying any-
one. Thus, there is a high requirement of giving an insight into Big Data Analysts
about the problems or threats related to privacy of the data prevailing currently and
proper involvement of such analysts is also expected for the development of much
better tools to protect and prevent such threats or issues.

5.4.4  Big Data Analytics in digital forensics

Basically, digital forensics comes under forensic science whose focus is on identi-
fication, acquisition, processing, and analysis followed by formation of the report
of the digital data stored. While Big Data is taken into consideration, there are few
points which act as a challenge in the case of digital forensics. Considering the
identification step, identifying the critical evidence out of humongous data available
collectively seems impossible. Out of lakhs of data, there is a high probability that
 Accurate management and progression of Big Data Analysis  109

only 10% of the data obtained is useful from a forensic point of view. Traditionally,
the data stayed in complete control of the forensic organisation itself but with the
use of Big Data, for all data processing, forensic experts have to depend on the
organisations to collect the data which simply increases the duration of analysis
and find the facts among them. In many case studies, it is found that for just col-
lecting and preserving the data, more than 24 hours are required. There are various
characteristics of the data, viz., volume, intensity, etc., which do not match with the
available methods in the present. So, examining and analysing such data in such a
programming paradigm becomes challenging. Keeping in mind all the challenges
faced in organising the data, special search and data mining techniques should be
employed to make it more efficient. After organisation, the last step involved in
digital forensics is presentation where the experts have to present the analysis or the
findings done from the evidence in the court. Using traditional computers to present
the analysis is easy compared to Big Data where the data is complex and providing
the analysis of such data becomes tedious for the jury to understand and evaluate.
Collectively, each step faces different challenges while using Big Data.
Thus, to face such challenges, a theoretical model of using Hadoop Distributed
File system (HDFS) and cloud-­based computing model is proposed. Hadoop or
Apache Hadoop is basically an open source software that utilises the network of
many computers to solve Big Data problems or huge data problems. To solve such
Big Data problems, it uses the MapReduce framework, which is already explained.
Coming to Hadoop Distributed File system, it is a file system component that is
developed in such a way that it can store huge data sets effectively and efficiently
and provides high bandwidth across the cluster for performing different applications
of the end user.
Apart from HDFS, cloud-­based computing models should be deployed in the
case where Big Data is considered as such data contain terabytes of information.
This architecture might prove to be cost effective and scalable.

5.5 Real-time edge analytics for Big Data in IoT

We all know that our tendencies are shifting from manual to autonomous devices.
These devices are connected in a way thereby creating a chain of appliances trans-
ferring/sharing data and other resources. These are called IoT where the connection
will be due to the internet/intranet/cloud. But if we keep all the processing in a sin-
gle processor then the load will be huge as well as the response will be slower. For
example, if we take an autonomous car then due to development in internet-­based
gadgets, e.g., traffic sensors, disseminated camcorders, and associated apparatuses,
a surge of information is being created, which is handled utilising edge-­registering
assets. The image of a vehicle in front of it or the traffic signals should be processed
faster to avoid the dangers of the road, and if we depend on cloud, e.g., then the
appliance should first transfer the data to the cloud where it will process the whole
information and send its reply in the form of a command to the respective func-
tional unit. This will take a lot of time compared to the case where we can give the
110  Intelligent network design driven by Big Data analytics

data required only for the command to the command unit. This can be achieved
by Edge Analytics. In this we process the data from the input at the source itself
thereby reducing the information transfer to the command unit thereby reducing the
response time.
The ascent of Big Data acquires uncommon new advantages and opens a few
application areas. The significant undertakings can misuse the IoT-­created informa-
tion and surmise the business esteem by performing information investigation. In
particular, a few applications would require continuous edge examination to rap-
idly discover helpful relationships, client inclinations, shrouded designs, and other
significant data that can help associations and chiefs to take increasingly educated
business activities. These difficulties speak to a few open doors for analysts in the
area to research various bearings including data combination, AI, and explanatory
instruments structure.
The study of IoT-­produced data is very important to meet the demands and
satisfy the consumers. The conventional investigation methods will not be suitable
due to the enormous data about various details like limits for various tasks like
registrations, timeframes for various tasks, etc. The IoT-­created Big Data has novel

Figure 5.10   Big Data in edge systems

 Accurate management and progression of Big Data Analysis  111

qualities, e.g., discontinuous clamour age, exceptionally unstructured, and dynamic

nature. These qualities make it difficult for constant investigation. The authorities
must take the issue of locally accessible and appropriate computational assets to
overcome the difficulty in continuously handling and examining the collected data.
The desire to improve the data handling has opened doors for interdisciplinary
research and innovation, e.g., Machine learning, AI, pattern recognition.
Big Data Analytics has a wide range of applications. Some of the major appli-
cations of Big Data Analytics can be understood from the following web chart in
Figure 5.10.
In banks and insurance companies, Big Data is very much helpful as it con-
nects the organisation directly to the repository of customer financial data. Due to its
capability to store large volumes of data securely, it has been proven really helpful
for the healthcare and research industries. Prior to this, all the historical data and
other important reports of patients, number of surgeries done per day in a hospital,
etc., were kept in handmade records in the record room where chances of losing the
record were quite high. With the advancement of Big Data, this problem has been
resolved to a great extent. In the case of retail industries, the connection between
users and large- and medium-­sized industries has been developed with the advance-
ment of Big Data. In the present scenario, social media is the link for all this and this
social media has grown with the development of Big Data.
Therefore, Big Data has a wide range of applications in the present scenario and
it is very helpful for accurate management of data.

5.6 Conclusion

With the knowledge of Big Data, we can minimise the effort for product enhance-
ment, and accumulation of new data can help in increasing user friendliness using
our understanding in logical programming. Logical programming has made the
entire information examination simpler. Establishing patterns reduces the time to
break down the enormous data in a cost-­effective pattern. They speak to a real
jump forward and an away from to acknowledge huge gains regarding proficiency,
efficiency, income, and gainfulness. Big Data Analysis is very important for proper
functioning in this digital era since everything is digital in nature. We need to be
careful about our data leakages and while notifying our personal data. With the
help and cooperation of both business and innovation experts, the age of Big Data
will reach greater heights. Besides all this, Big Data Analytics is really helpful for
medium and large organisations to connect with users all over the world through
social media.
Transcript from one of the most alluring speakers on the Big Data circuit,
Google’s Avinash Kaushik, from his introduction at [12], ‘A Big Data Imperative:
Driving Big Action’:
‘I really don’t generally think about the guarantee of information except if they
can convey on that guarantee that accompanies the information’.
112  Intelligent network design driven by Big Data analytics

References

[1] HongJu X., Fei W., FenMei W., XiuZhen W. ‘Some key problems of data man-
agement in army data engineering based on big data’. IEEE 2nd International
Conference on Big Data Analysis (ICBDA); Beijing, China, IEEE, 2017. pp.
149–52.
[2] Sagiroglu S., Sinanc D. ‘Big data: a review’. International Conference on
Collaboration Technologies and Systems (CTS); San Diego, CA, IEEE, 2013.
pp. 42–47.
[3] Shalaginov A., Johnsen J.W., Franke K. ‘Cyber-­crime investigations in the
era of big data’. IEEE International Conference on Big Data (Big Data);
Boston, MA, IEEE, 2017. pp. 3672–76.
[4] Jo J., Joo I., Lee K. ‘Constructing national geospatial big data platform: cur-
rent status and future direction’. IEEE 5th World Forum on Internet of Things
(WF-­IoT); Limerick, Ireland, IEEE, 2019. pp. 979–82.
[5] Lee C.K.M., Yeung C.L., Cheng M.N. ‘Research on iot based cyber physi-
cal system for industrial big data analytics’. IEEE International Conference
on Industrial Engineering and Engineering Management (IEEM); Singapore,
IEEE, 2015. pp. 1855–59.
[6] Mazumdar S., Wang J. ‘Big data and cyber security: a visual analytics per-
spective’ in Parkinson S., Crampton A., Hill R. (eds.). Guide to Vulnerability
Analysis for Computer Networks and Systems. Computer Communications
and Networks. Cham: Springer; 2018.
[7] Terzi D.S., Terzi R., Sagiroglu S. ‘Big data analytics for network anomaly
detection from netflow data’. International Conference on Computer Science
and Engineering (UBMK); Antalya, Turkey, IEEE, 2017. pp. 592–97.
[8] Azvine B., Jones A.. (eds.) ‘Meeting the future challenges in cyber securi-
ty’ in Dastbaz M., Cochrane P. (eds.). Industry 4.0 and Engineering for a
Sustainable Future. Cham: Springer; 2019.
[9] Priyadarshini S.B.B., BhusanBagjadab A., Mishra B.K. ‘The role of IoT and
big data in modern technological arena: a comprehensive study’ in Balas
V., Solanki V., Kumar R., Khari M. (eds.). Internet of Things and Big Data
Analytics for Smart Generation. Intelligent Systems Reference Library. 154.
Cham: Springer; 2019.
[10] Puthal D., Ranjan R., Nepal S., Chen J. ‘IoT and Big Data: An Architecture
with Data Flow and Security Issues’ in Longo A. (ed.). Cloud Infrastructures,
Services, and IoT Systems for Smart Cities. IISSC 2017, CN4IoT 2017. 189.
Springer, Cham; 2018.
[11] Praveena M.D.A., Bharathi B. ‘A survey paper on Big Data analytics’.
International Conference on Information Communication and Embedded
Systems (ICICES), Publisher: IEEE; Chennai, India; 2017. pp. 1–9 pp..
[12] Singh S., Singh N. ‘Big Data analytics’. 2012 International Conference on
Communication, Information & Computing Technology (ICCICT); 2012. pp.
1–4.
 Accurate management and progression of Big Data Analysis  113

[13] Jayasingh B.B., Patra M.R., Mahesh D.B. ‘Security issues and challenges
of big data analytics and visualization’. 2nd International Conference on
Contemporary Computing and Informatics (IC3I); Greater Noida, India, 14–
17 Dec; 2016. pp. 204–8.
[14] Gupta B., Kumar A., Dwivedi R.K. ‘Big data and its applications– a review’.
2018 International Conference on Advances in Computing, Communication
Control and Networking (ICACCCN); Greater Noida, India, IEEE, 2018. pp.
146–49.
[15] Haidar M., Kumar S. ‘Smart healthcare system for biomedical and health care
applications using aadhaar and blockchain’. 5th International Conference on
Information Systems and Computer Networks (ISCON); Mathura, India, 22-­
23 Oct 2021; 2022. pp. 1–5.
[16] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
Personal Communications. 2021, vol. 10(3), pp. 1–14.
[17] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[18] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[19] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering (IJITEE). 2019, vol. 8(6), pp.
1672–6.
[20] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[21] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[22] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road Surface Quality Monitoring
Using Machine Learning Algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[23] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and IoT for
smart road traffic management system’. Proceedings of the IEEE India
Council International Subsections Conference (INDISCON); Visakhapatnam,
India, 3-­4 Oct; 2020. pp. 289–96.
[24] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation
of fault tolerance technique for internet of things (iot)’. 12th International
114  Intelligent network design driven by Big Data analytics

Conference on Computational Intelligence and Communication Networks

(CICN); Bhimtal, India, IEEE, 2020. pp. 154–59.
[25] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. Proceedings of the
Confluence 2020 - 10th International Conference on Cloud Computing, Data
Science and Engineering; Noida, India, IEEE, 2020. pp. 63–76.
[26] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. Proceedings of the
Confluence 2020-­10th International Conference on Cloud Computing Data
Science and Engineering; Noida, India, IEEE, 2020. pp. 63–76.
[27] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. IEEE; Mathura, India, 2019. pp. 250–55.
[28] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
Proceedings of the International Conference on Computational Intelligence
and Communication Networks; Jabalpur, India, IEEE, 2016. pp. 79–84.
[29] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Progress in electromag-
netics research Symposium’. A Utility Maximization Approach to MAC Layer
Channel Access and Forwarding. 2015, vol. 2015, pp. 2363–7.
[30] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wire-
less sensor network’. Proceedings of the 5th International Conference on
Communication Systems and Network Technologies CSNT; Gwalior, India,
IEEE, 2015. pp. 194–200.

[31] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distribut-
ed localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, 7-­8 Feb; 2014.
[32] Sudhakaran S., Kumar S., Ranjan P. ‘Blockchain-­Based Transparent and
Secure Decentralized Algorithm’. International Conference on Intelligent
Computing and Smart Communication 2019. Algorithms for Intelligent
Systems; Singapore: Springer; 2020.
[33] Kumar S., Trivedi M.C., Ranjan P. Evolution of Software-­Defined Networking
Foundations for IoT and 5G Mobile Networks. Hershey, PA: IGI Publisher;
2020. p. 350.
[34] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India, IEEE, 2021. pp. 1–6.
 Chapter 6
Cram on data recovery and backup cloud
computing techniques
Dharanyadevi P 1, Julie Therese M 2, Senthilnayaki B 3,
Devi A 4, and Venkatalakshmi K 5

The present digital world technology is evolving at a rapid pace. To store, man-
age and protect the digital information, it is necessary to back up and recover the
data with utmost efficiency. As a solution, cloud computing that offers customers a
wide range of services can be used. Storage-­as-­a-­Service (SaaS) is one of the cloud
platform’s services, in which a large volume of digital data is maintained in the
cloud database. Enterprise’s most sensitive data are stored in the cloud, ensuring
that it is secure and accessible at all times and from all locations. At times, informa-
tion may become unavailable due to natural disasters such as windstorms, rainfall,
earthquakes, or any technical fault and accidental deletion. To ensure data security
and availability under such circumstances, it is vital to have a good understanding
of the data backup and recovery strategies. This chapter examines a variety of cloud
computing backup and recovery techniques.

6.1 Introduction
Cloud computing is an idea of affording on-­demand network access to a communal
pool of configurable computing resources (e.g., network, servers, applications, ser-
vices and storage) that may be swiftly supplied and released with no service provider

1
Department of Computer Science and Engineering, Puducherry Technological University, Puducherry,
India
2
Department of Electronics and Communication Engineering, Sri Manakula Vinayagar Engineering
College, Puducherry, India
3
Department of Information Science and Technology, College of Engineering, Anna University,
Chennai, India
4
Department of Electronics and Communication Engineering, IFET College of Engineering,
Villupuram, India
5
Electronics and Communication Engineering, University College of Engineering Tindivanam,
Tindivanam, India
116  Intelligent network design driven by Big Data analytics

or administrative effort contact. The cloud computing affords the computing ser-
vices via the Internet in order to afford more elastic resources. The major advantage
of using cloud services is that we have to pay only for the services we use, which
helps to run the organization efficiently in terms of economical scale [1].

6.1.1  Origin of cloud

• “Cloud computing” has its origin in the untimely days of the Internet, when we
drew the network as a cloud [1].
• “First cloud around networking (TCP/IP abstraction), which was hidden from
us by the cloud”—Kevin Marks, Google.
• A second cloud has formed around the documents (WWW data abstraction).
• The cloud abstracts the complexities of servers, data, apps, and assorted plat-
forms from the infrastructure (—Jeff Bezos, Amazon CEO).

6.1.2  Sole features of cloud computing

• Shared and extensible responsibility
• Outsourcing data and applications
• Virtualization and hypervisors
• Multi-­tenancy
• Service-­level agreements
• Compliance and regulations
• Heterogeneity

Cloud computing has been hailed as the definitive and far-­reaching solution
to IT companies’ growing storage expenses. Cloud computing has emerged as one
of the most popular strategies for delivering IT services. Clouds have arisen as a
computing infrastructure that enables the dynamically scalable, virtualized supply
of computing resources as a utility. IT behemoths including Amazon, Google, IBM
and Microsoft have all launched cloud computing initiatives. Because of the finan-
cial benefits, data outsourcing to cloud storage servers is becoming more popular
among businesses and customers.

6.1.3  Advantages of cloud computing

• Easy implementation
• Accessibility
• No hardware required
• Efficient recovery
• Flexible for growth
• Cost per head
 Cram on data recovery and backup cloud computing techniques  117

Table 6.1   Types of storage

Parameters Hot storage Cold storage Warm storage

Characteristic Very fast access Fast access Medium speed access
Application Data storage Cloud services Remote server
Examples Hard drives, flash Cloud drives, Google Cheaper and larger
drives and SSDs cloud storage and spinning drives
Amazon Web
Services (AWS)

6.1.4  Disadvantages of cloud computing

• Bandwidth issues
• No redundancy
• No longer in control

Storage devices created primarily for cloud-­based provisioning are represented

by the cloud storage device mechanism. In order to enable the pay-­per-­use method,
cloud storage devices are frequently equipped to provide fixed-­increment capac-
ity allotment. The security, integrity and secrecy of cloud storage are the major
concerns. Many larger organizations use cloud storage to enfold their information.
Cloud storage is mostly used for backing up user’s data. Due to a large backup of
stored data, cloud storage struggles with several issues such as efficiency, storage
and security problems. The three types of storages are hot storage, warm storage and
cold storage. The differences between the three storages are illustrated in Table 6.1.
Approximately 2.5 quintillions of information are produced every day, as per the
survey conducted by Vouchercloud. In 2013, the data traffic is 28 875 GB/s accord-
ing to the survey, which then rose to 50 000 GB/s by 2018 [2]. In the cloud storage
environment, the greatest obstacle is to handle larger data and sustain an enhanced
information generation rate. However, in different language, we can say that the
cloud offers services that can be classified into three groups such as Infrastructure as
a service (IaaS), Platform as a service and Software as a service.
Cloud computing is now being used as a backup resource for storing data [1, 3].
A service-­oriented business model is used in cloud computing. Data backup is very
essential because of the high chances of hardware failure, software errors or human
mistakes, which lead to the loss of confidential information. Because of the greater
risk for failures or theft, backup of data is more essential than ever for devices like
desktop computers, laptops and mobile phones.
Smartphones are generally more widely used as computing devices, so a great
deal of data is stored on those devices [4]. These reasons make it more important to
store data in the cloud and hence avoid failures or security threats on the devices.
Due to the rapid technical development, smart devices make data recovery more
difficult to recover the saved data from the cloud; henceforth, mobile devices face
challenges in backup and retrieval of data in a cloud environment.
118  Intelligent network design driven by Big Data analytics

Data recovery is the process of restoring the lost data due to data exploitation
or deletion unknowingly, or the data not being accessible due to technical reasons.
The restoration process is done from the system where the information was already
backed up.
The types of data that you certainly should be backing up may include:

1. Any files: word documents and excel spreadsheets are common examples of
information-­carrying files that must be backed up. The document may also be
emailed to a mailbox or uploaded to a cloud service, allowing it to be accessed
at any time and from any place.
2. Any important entries made during the day should be secured and copied.
Emails, calendar entries, browsing data, contacts, social media records and
some unnoticed applications are some of the entries.
3. The media files such as photographs, images, audio and video. Media files are
usually much bigger, necessitating a different approach while storing in the
cloud.

Here are a few significantly relevant backup technique recommendations: First

and foremost, any backup containing confidential information should be encrypted.
Second, copies can be stored both in offline and online locations. Third, always
double-­check the backups to ensure that the data can be recovered at ease.
Restore data contained in deleted or corrupted files in full or in part. If file dele-
tion is not exceeded (or the disc is not formatted after deletion), the area enclosed
by the deleted file will be retrievable via popular software. In case of corrupted files
(or stored on damaged media), customized software and specialized technology are
needed for the recovery of damaged files. This chapter addresses the study of quan-
titative analysis on data recovery and backup techniques. The cloud storage is equal
to data and cloud computing is equal to processes.
The chapter is organized as follows: Section 6.2 addresses the various recovery
and backup techniques, and Section 6.3 discusses a thorough overview of the survey
on different data recovery and backup technologies used in the cloud as well as their
benefits and drawbacks. Section 6.4 addresses the chapter’s conclusion and it is
backed by references.

6.2 Classification of data recovery and backup

6.2.1  Recovery
The data recovery services are offered to collect data by recovery techniques.
Different companies offer different strategies. The data recovery techniques used in
the following devices are:
Hard Drive Recovery: A significant number of data recovery services happen
due to hard disc failures. Due to the advanced technology, hard drives are grow
larger, and more data retrieval is needed than ever, as users or customers are con-
cerned about losing their data.
 Cram on data recovery and backup cloud computing techniques  119

Optical Recovery: Optical media are an optical device that writes or reads data,
e.g., CDs and DVDs. Many complications can lead to a failure of optical media: the
CD or DVD can be scratched; the CD or DVD player can destroy the media.
Removable Recovery: These removable storage devices have a high chance to
erase the data than the other media platforms already discussed; hence, they are less
incurred.
RAID Recovery: It is really difficult to have Redundant Array of Inexpensive
Devices (RAID) systems because of the extreme skills required to build, manage
and maintain. Fault and configuration problems are more likely to occur. If RAID is
not backed up correctly, business failure or severe loss can be incurred.
Tape Recovery: Tape is used mostly for backup successive storage media; it
uses huge backup space. Experts having distinctive technologies and software tools
use tape as a recovery device.
Digital Recovery: Cameras, handheld storage devices and flash media are
included in the digital media. Since they are cheaper, a growing market for many
goods has been created for recovery. Data recovery organizations have introduced
suitable solutions for digital data loss.

6.2.2  Backup
Laxmi et al. explained that cloud backup is a kind of service used by the cloud to create,
edit, maintain and re-­establish enterprise cloud resources and infrastructure [5]. This is
achieved through the internet remotely. Both online backup and remote backup can be
called as cloud backup. In cloud computing, the online backup via the internet can be used
anywhere at any time [6]. Online backup can be used without paying much importance
to technical/physical management that includes HSDR (High-­Speed Data Rate Transfer),
PCS (Parity Cloud Service), ERGOT (Efficient Routing Grounded on Taxonomy) and
Linux Box.

Advantages
• It delivers high performance through parallel use.
• The PCS is very quick and easy to use. Further versatile for data retrieval based
on the parity retrieval service.

Disadvantages
• High access to the Internet: To retrieve the data, it requires fast Internet speed.
The users cannot access the files in case of network problems.
• Data are held on the server of third parties: Though backup of the main server
to an external server seems to be a good idea, it can also be a disadvantage. If
something goes wrong with the external server, users lose the data.
• Determine the bandwidth grant: While some service providers allow unre-
stricted bandwidth, many other providers provide a restricted allowance and
can charge an extra fee when users exceed their allocation.
120  Intelligent network design driven by Big Data analytics

Figure 6.1   Full backup strategy

The three major categories of backup services are full backup, differential
backup and incremental backup. The definitions of three types of backup techniques
are as follows:
Full backup: A complete backup or full backup is shown in Figure 6.1. The full
backup is the most essential and primary kind of backup process. This category of
backup copies all details into a different media package. The benefits of full backup
are it can be carried out after each operation and a complete copy of all data will be
available within a single medium.
Differential backup: The strategy used in differential backup is shown in
Figure 6.2. It copies all modified results from the last backup. However, every time
it is executed, it continues to copy all changed data and it takes additional space and
time to complete the previous complete backup.
Incremental Backup: It is a gradual backup process that could only duplicate
customer data that has changed since the last backup. The benefit of an incremental
backup is that it duplicates only a few records and hence consumes less time and
space. By this way, this operation is faster and requires less memory necessary for
backup savings; this process is shown in Figure 6.3.

6.3 Study on data recovery and backup cloud computing

techniques

6.3.1  Backup of real-time data and recovery using cloud computing

Karishma et al. proposed backup of real-­time data and recovery using cloud comput-
ing. The author explained that the data backup and recovery are essential to manage
the data expeditiously. The author has proposed Seed Block Algorithm (SBA). The
SBA is an efficient technique for data recovery. First and foremost, the SBA algo-
rithm is proposed to decrease the recovery time. In addition, the algorithm performs
two tasks during the recovery phase [7]. In the first task, users can collect data from

Figure 6.2   Differential backup strategy

 Cram on data recovery and backup cloud computing techniques  121

Figure 6.3   Incremental backup strategy

remote locations without the network connection. Second, if the cloud storage gets
debased, which leads to loss of data, the user can recover the data easily. It also
focuses on the data security stored on the remote server. The working mechanism of
the SBA technique is as follows:

• As shown in Figure 6.4, the remote data recovery server, main cloud and the
number of consumers are the components of the SBA.
• The first step is to establish random numbers for each client in the central cloud
as a unique user ID.
• The second step is to create a seed block value for a user. For this, exclusively-
­OR (XOR) operation is executed between the user ID and the random number
when the user is trying to access the main cloud so that the user ID is being
recorded in the cloud.
• When the user in the main cloud generates a new file, the seed block value of
that client is stored in the remote data backup server.
• The folder of the user remains XORed with the seed block value of the same
user in the cloud server.
• The resulted XORed file gets stored in the remote data backup server as a file.
• If unlucky, the file which is in the main cloud gets corrupted or deleted, the user
can access his/her file by doing an XOR operation between the file and the seed
block value of that respective user.

The technologies used in SBA techniques are as follows:

Figure 6.4   Real-­time data backup and recovery

122  Intelligent network design driven by Big Data analytics

• High-­security distribution and rake technology (HS-­DRT): HS-­DRT technol-

ogy is used in devices like tablets and smartphones. But its implementation cost
is high and has less redundancy.
• Parity cloud services: This technique provides reliability and privacy.
Additionally, the cost of implementation is low.
• Efficient routing grounded on taxonomy (ERGOT): This technique also pro-
vides privacy. But it is more complex to implement and takes more time to
recover the data.
• Linux box: The Linux box is a computer running a version of LINUX, simple
and easy to implement. Yet it requires more bandwidth.
• Cold/Hot backup strategy: This technique helps only during failures. At the
same time, the cost of implementation increases as the size of data increases.

Advantages
• SBA takes less time to perform the recovery operation.
• It provides data integrity and security.
• It also ensures reliability of data.
• The implementation of this algorithm is easy and also a cost-­efficient method.
• When the main cloud could not provide the requested data to the user, it aids in
recovering the lost data from the remote location.

Disadvantages
• The cost of implementation increases as the size of data increases.
• SBA takes more time to recover the data.

6.3.2  Data recovery and security in cloud

Jayachander et al. explained that cloud computing is very popular because of the
rapid number of services given to an individual user and the organization as a whole
[8]. Data recovery is the main concept of cloud computing. Using data recovery tools,
the one who has the access to infrastructure can recover the files even if it was deleted
from the cloud. So data recovery can lead to security issues. The author has intro-
duced rename method. The rename method is used to meet the security concern, it
is used to protect information and it works by renaming all files. It changes the file
format; if the file is in JPEG then the file format is changed to mp3 by the rename
module. This does not delete the original file, but without understanding the actual
data format, prediction of the original data is impossible. Many who have access to
infrastructure cannot retrieve the content without understanding it. Without fear of
recovery, data can be removed in the cloud [9].
 Cram on data recovery and backup cloud computing techniques  123

Advantages
• PhotoRec application is used to recover the data even if the data were removed
from the memory location.
• Using memory analysis software, the deleted data from the cloud can be
retrieved after deletion from the cloud.
• If the confidential records about users are erased by mistake, they can be
recovered.
• Rename module gives high protection to the user’s information.
• Rename changes the format of the original data type so data cannot be predicted
or hacked by anyone.
• The information contained in the data remains understandable unless the exact
type of data is known; as the original data format is difficult to know, data can-
not be retrieved.

Disadvantages
• Security problems are created with data recovery tools.
• Cloud servers recover the private data of users even if they delete the data from
the cloud.
• Reconstructing the user’s data results in compromise of privacy.

6.3.3  A
 moeba: An autonomous backup and recovery solid-state
drives for ransomware attack defense
Donghyun et al. explained that ransomware is a type of malicious software that
prevents users from using their devices and other computing resources unless they
pay a ransom. Because ransomware can cause rapid economic loss, an efficient tech-
nique of ransomware mitigation is imperative [6]. Existing methods for detecting
and avoiding ransomware rely on recognizing known ransomware behaviors. In the
existing methods, the flash guard and SSD (Solid-­State Drives) –- Insider is used for
data backup. In this chapter, Amoeba is proposed to solve the problem of ransom-
ware affecting the SSD backup and retrieval process. Amoeba instinctively executes
ransomware intrusion detection, alerting the entire system, data backup and retrieval
within the SSD.
In the proposed method, for all page write activity, Amoeba measures the prob-
ability of ransomware attacks. In the Ransomware Attack Risk Indicator (RARI)
hardware module, the Direct Memory Access (DMA) controller is expanded to
include a NAND flash inside the SSD. The RARI module calculates all page’s risks
and is used to decide whether the page is to be backed up.
Ransomware Risk Calculation: Intensity, similarity and entropy of Amoeba’s
RARI are used to identify possible ransomware attack risk indicators. By counting
the number of written requests, the intensity of the attack can easily be measured.
This is supported by Flash Translation Layer (FTL) firmware which is introduced
124  Intelligent network design driven by Big Data analytics

at a low price. The DMA controller gets access to new and old data to measure
similarities between the data. Therefore, it must issue the NAND flash memory with
an extra page, which is read, to store older data. Fresh information, old information
from the flash memory and the re-­assembly of the controller are read from the main
memory. Amoeba uses a powerful classifier of ransomware that shows entropy, sim-
ilarity and strength. The RARI value is produced by standardizing the three indica-
tors using the new methodology named MinMaxScaler and carrying out the logistics
categorization. The following equation can be used to formulate the RARI value:
1
= ,  = ˛  SIM + ˇ  ENT +   INT + ı (6.1)
‍ 1+ ‍
Here in (6.1), the result of linear equations is given by ‍ ‍; the mean entropy, similarity
and write intensity are given by SIM,ENT, INT; the mean weights are ‍˛, ˇ,  ‍ and
‍ı‍stands for bias.
Backup and Recovery: To incorporate the recovery and backup function,
Amoeba utilizes the Out-­Of-­Band (OOB) section of a page. The page’s OOB seg-
ment includes the RARI value and Backup Page Number. The backup technique
initially tests whether the write request follows the read-­after-­write pattern for every
overwrite request. Subsequently, it verifies if the present valid page RARI value is
greater than the threshold value specified. Amoeba would consider it as a written
request of ransomware attack if both the requirements are met. If one of them is not
met, it is considered as a regular write request.

Advantages
• For quick ransomware detection, amoeba inserts special RARI computing
equipment in the DMA module.
• It autonomously backs ups and restores the computer.
• RARI is an implementer using TFL at a low cost.

Disadvantages
• The detection of ransomware attacks and the management of backup data from
any device have limited efficiency.
• Consequently, low-­level backup pages, which have a low likelihood of being
infected with ransomware, are invalidated by the SSD, when the backup state
page space becomes too large. It limits the backup page size by stabilizing the
SSD’s performance.

6.3.4  A
 cloud-based automatic recovery and backup system for
video compression
Megha et al. explained that cloud computing provides data recovery and backup effi-
ciently. It is the technique of storing the data or services in a remote place where one can
 Cram on data recovery and backup cloud computing techniques  125

gain access to it from anywhere and at any time [10]. Compression techniques are an
integral part in cloud computing, as it decreases the bandwidth required for storing and
accessing information from the cloud. Loading data takes place via both local backup
as well as cloud backup. In the local backup, compression takes place for the video files
with the help of FFMPEG (Fast Forward MPEG (Motion Picture Experts Group)) [11].
The cloud backup takes place in Microsoft one drive, where the graphs of compressed
and uncompressed files are examined for reduced bandwidth. And the user alone can
access the data from the cloud at any time from anywhere across the world. Thus, two
backups are carried out which will help in the protection of data along with compression
techniques for reduced bandwidth. The core objective of the chapter is to provide two
backups and video compression in the cloud.
The existing system of Dropbox storage does not have a recovery methodol-
ogy. If the data are lost in the local system, it will automatically get lost from the
Dropbox. Secondly, to store heavy files like audio and video files into the Dropbox,
lots of internet data are needed. This makes it difficult to back up and recover data
for a user. Hence moving to cloud-­based recovery and backup which uses a Dropbox
system with automatic recovery is vital. The video files would get compressed in
the background and stored in the cloud. The background entities involved in the
proposed system are as follows:

• Skydrive API’s Library: It is provided by the Microsoft cloud services for stor-
ing a large amount of data. It is also called as OneDrive or SkyNet.
• FFMPEG: It is a multi-­media framework used for the compression of video
and audio files.
• Cloud Storage as a Service: It is nothing but providing storage of huge amount
of data as a service in the cloud.
• File System Tracker: It is a web-­based program for monitoring and transferring
items.

To communicate with the cloud from the system, .NET is an efficient lan-
guage for the browser. Three methods must automatically carry out simultane-
ously in the local storage as well as cloud storage. They are automatic backup,
recovery system and video compression. The system performs pre-­backup check
daily.

Advantages
• Cloud-­based recovery and backup uses a Dropbox system with automatic recov-
ery. It helps in recovering data from the cloud storage even though the data have
been deleted from the local storage. This is called as dual backup.
• The system also provides video compression with the help of libx264 encoder
at the back group automatically before backing them up into the cloud storage.
• The system also provides high reliability of user data.
126  Intelligent network design driven by Big Data analytics

Figure 6.5   Three modules of EGA

6.3.5  E
 fficient and reliable data recovery techniques in cloud
computing
Praveen et al. described that cloud computing provides various computing services
for global communication; one such service is SaaS [12]. Sensitive and essential
data are stored in the cloud in a remote site and can be accessed from anywhere
at any time. And if there is a failure or disaster, these data should be kept secure.
For this, data backup and recovery methods should be used to preserve the level of
security [13].
Many methods exist for recovering data from the cloud, but most of them are
inefficient and unreliable. A system with three modules has been proposed and it is
named as an Enriched Genetic Algorithm (EGA) to meet efficiency and reliability
in the data recovery process. The modules are user module, remote backup server
and main cloud server, as shown in Figure 6.5. Data are stored in both main cloud
server and remote server to meet efficiency, and to provide reliability, more backup
servers are used [14–17].
The procedure followed in EGA is as follows:

• User uploads file to the main cloud server and the number of backup cloud
servers.
• Hash code H1 of the file is generated and it is stored in the database.
• The number of copies of replicas is selected for storage.
• The size of the file is calculated.
• For downloading the files, the user has to select the file and download it. If it
is not in the main cloud, then it should be recovered from the backup storage.
• For recovering file, hash function H2 is generated.
• If both hash functions H1 and H2 are equal, then the original file is recovered.
 Cram on data recovery and backup cloud computing techniques  127

Advantages
• If the data are lost in the main server, it can be recovered from the cloud server.
• Provides reliability by creating a number of backup servers.
• Multi-­server system increases the data availability.
• Hashing function like MD5 is generated to provide integrity.
• Uploaded either in two-­block or four-­block server.
• Four-­block server provides flexibility, i.e., lost files can be recovered from any
backup servers.

Disadvantages
• Requires more storage space.
• Four-­block server takes more upload and recovery time than two-­block servers.
• A replica of the same file is created N times.

6.3.6  Cost-efficient remote backup services for enterprise cloud

Yu et al. explained that the backup process creates large transmission overheads and
potential faults as a result of long-­distance and expensive network communication.
The author proposes Neptune, a well-­organized remote communication service, as
a solution to this issue. Neptune sends enormous information dynamically between
long-­haul data centers. The cost-­effective filtration structure makes it easier. The
Neptune filtration is seen as ignoring consistency and compacting data similarity as
clouds are composed of duplicate information. In this chapter, the data filtration uses
two main techniques: chunk-­level de-­duplication and delta compression. The objec-
tive of these two techniques is to avoid the transmission of redundant data.
The de-­duplication scheme divides the files into a large number of chunks. Each
of which is individually recognized by a hash signature and is known as a finger-
print. Duplicate chunks must be ignored when examining their fingerprints. In addi-
tion, delta compression compresses the same regions by calculating the variations
between them. Exploiting and exploring the features of redundancy increase the
efficiency of remote backups [18, 19]. In a rational environment for an effective web
redundancy exclusion scheme, the Neptune is feasible. It also increases the use of
network resources availability.
Data Filtration: As already mentioned, there are two categories of data filtration:

1. Chunk level de-­duplication

2. Data compression

The de-­duplication is a vital part of a modern backup that separates data pack-
ets into variable-­length chunks and afterward substitutes duplicate chunks for their
pre-­stored samples with pointers. A de-­duplication method can recognize each
chunk via its hash code, i.e. fingerprint. However, in reality, due to various lengths
of chunks, the de-­duplication method handles data on a bigger entity known as a
128  Intelligent network design driven by Big Data analytics

container. A container is fixed in size and is commonly used as reading and writing
units. To secure the location of the network traffic, Neptune adds the chunks into
the containers. Neptune often utilizes the container as a pre-­fetching device. The
Least Recently Used (LRU) algorithm is used to remove the preserved cached in a
container.
The second one is a rapid and more powerful design delta compressor. It exploits
file correlations and can generate compressed files drastically in a smaller size. To
enhance network transmission, therefore, it takes considerably smaller bits to trans-
mit; just the difference (or delta) between two data is required. A compressor that
allows two inputs is used in delta compression, where one input is the destination
file that has to be compressed and the second is a source file used for reference. The
creator of the delta compressor identifies and replicates the discrepancies between
both the destination and the source files. The delta and source data are used to pro-
duce an accurate replica of the destination from the compressor.
Delta Compression and Recovery: It is suggested that a delta-­ based de-­
compression scheme can calculate the distinction between several identical files
besides the original one to provide an economical and effective recovery solution.
The users can artificially pick the base file or automatically by Neptune, which uses
well-­known and powerful clustering algorithms. Neptune requires only maintaining
the deltas from other de-­duplicated files and the base file. The delta-­based archi-
tecture will reduce consumption of device resources dramatically and enable data
recovery functionalities. Neptune stores all the base files even if they are chosen for
retrieval by another user in order to promote the data retrieval during data recovery.
Neptune performs delta decoding procedures by computing the database files and
the deltas when customers want to retrieve information.

Advantages
• The most important idea of the chapter is to detect similarities, delta chains
along with compression and restoration, and to evaluate performance using
industrial datasets in the modern world.
• It provides remote backups at a low cost.
• It does not allow additional calculation on the deltas which reduces the latency.
• Neptune often uses shortcut delta chains to help quick data recovery.

Disadvantage
• Time consumption is high.
• Due to the complexity of the procedures, financial inefficiency is likely to occur.

6.3.7  DR-cloud: Multi-cloud-based disaster recovery service

Yu et al. explained Cloud Disaster Recovery (CDR). CDR is a technology that
allows remote computers to be recovered and backed up in a cloud-­based milieu.
 Cram on data recovery and backup cloud computing techniques  129

Figure 6.6   Disaster recovery in multi-­cloud environment

CDR is predominantly an IaaS approach that backs up and designates the data in
Remote Off-­site Cloud Server database. IT systems are being used by an increasing
number of providers, and some of them are financial and non-­financial services.
Many significant businesses and government services use the process of disaster
management to protect key data and to reduce downtime triggered by disaster device
defects [19–33].
The different types of methods used in CDR are regular backup or persistent
information synchronization and preparation for a standby network in geographi-
cally isolated areas [34–37]. Figure  6.6 shows the disaster recovery in the multi-­
cloud environment.

Advantages
• Using multiple clouds not only allows picking the most cost-­effective options
but it also allows choosing the best cloud services to fulfill specific business
needs.
• Adopting a multi-­cloud strategy can also enable businesses to avoid vendor
lock-­in decreasing their dependency on a single cloud supporter.
• Using the retrieval of multi-­cloud catastrophes, one can duplicate their resources
to a secondary cloud provider in some other location.
• Respective cloud provider’s Disaster Recovery (DR) services are planned to
deal with numerous cloud suppliers.

Disadvantages
• If the system is not managed correctly, the costs of the use of this system may
increase and the business agility will be affected.
• A secured approach and a multi-­cloud approach are the crucial areas to be han-
dled effectively.
Table 6.2   Data recovery and backup cloud computing techniques

Title Techniques Pros Cons

Backing up of real-­time data and Seed block algorithm • It takes less time to perform the • The cost of implementation increases
recovery using cloud computing recovery. as the size of data increases.
• It ensures reliability. • SBA takes more time to recover the
data.
Data recovery and security in the Rename method The data can be recovered even it Security and privacy problems are
cloud is removed from the memory created.
locations.
An autonomous backup and Amoeba • Autonomous computer backup It limits the space of the backup sites.
recovery SSD for ransomware and restoration.
attack defense • Negligible at cost.
A cloud-­based automatic recovery It uses a Dropbox system with • Provides better bandwidth. • Security is less.
and backup system with video automatic recovery • Also provides high reliability • Time consumption is high.
compression of data.
Efficient and reliable data recovery EGA • Multi-­server system increases • Requires more storage space.
technique in cloud computing the data availability. • Replicas of the same file are created
• Provide reliability by creating N times.
more number of backup servers.
Cost-­efficient remote backup servers Neptune • Cost-­effective remote backups. • Time consumption is high.
for enterprise clouds • Reduces the amount of • Due to the complexity of the
computation time it takes for the procedures, the financial inefficiency
intermediate deltas. is likely to occur.

DR-­Cloud: Multi-­cloud-­based DR-­Model cloud • Allows replication of resources This methodology is made more
disaster recovery to a different geographic region difficult due to complicated security
using a second cloud provider. services and a multi-­cloud approach.
 Cram on data recovery and backup cloud computing techniques  131

• The major problem in multi-­cloud is handling, reliability, consistency, and high

accessibility with most service providers.
• In situations where data recovery is not handled effectively, financial ineffi-
ciency is likely to occur.

Table 6.2 illustrates the data recovery and backup and cloud computing tech-
niques. The table explains the techniques, advantages and disadvantages,

6.4 Conclusion

The cloud provides an advancement of using internet in different ways. It boosts

many new technologies on the internet. Cloud is basically computing as a service.
All the services are used by a broad range of cloud service consumer with differ-
ent classes of devices. Cloud works on pay-­as-­you-­go, where you pay only for the
resource used. The major use of the cloud is for data storage, data centers for a
backup or availability purpose. The data have to travel a long distance and pass
through many different networks before reaching the data centers. This chapter
addresses the various data backup and recovery techniques used in cloud computing
with its advantages and disadvantages. The backup techniques focus on efficiency,
reliability, implementation costs and minimum complexity. The best way to save the
data is backing them up in online, i.e., cloud storage. The system may face many
ransomware attacks or disasters; it is possible to access data from the cloud by main-
taining its privacy. This chapter also discusses various recovery techniques that help
the users in case of no data backup availability.

References

[1] Dharanyadevi P., Venkatalakshmi K. ‘Proficient selection of gateway and base

station by adroit algorithm in cloud-­VMesh network’. International Journal
of Communication Systems. 2016, vol. 30(6).
[2] Dihuni. Every Day Big Data Statistics - 2.5 Quintillion Bytes of Data Created
Daily. Dihuni [online]. 2020. Available from https://www.dihuni.com/2020/​
04/10/every-day-big-data-statistics-2-5-quintillion-bytes-of-data-created-​
daily/ [Accessed 9 Jun 2021].
[3] Kumar M., Sujatha P., Dharanyadevi P. ‘A detailed study with novel approach
for Extenuating DDoS in cloud computing’. Mitigating EDoS in Cloud
Computing Using In-­Cloud Scrubber Service. LAP LAMBERT Academic
Publishing. 2012.
[4] Levitin G., Xing L., Dai Y. ‘Optimal Backup Distribution in 1-­out-­of- ${N}$
Cold Standby Systems’. IEEE transactions on systems, man, and cybernetics.
Systems. 2015, vol. 45(4), pp. 636–46.
[5] Therese M.J., Dharanyadevi P., Harshithaa K. ‘Integrating IoT and cloud
computing for wireless sensor network applications’. Cloud and IoT-­Based
Vehicular Ad Hoc Networks. 2021, pp. 125–43.
132  Intelligent network design driven by Big Data analytics

[6] Min D., Park D., Ahn J., et al. ‘Amoeba: an autonomous backup and recovery
SSD for Ransomware attack defense’. IEEE Computer Architecture Letters.
2018, vol. 17(2), pp. 245–8.
[7] Monisha S., Venkateshkumar D.S. ‘Cloud computing in data backup and
data recovery’. International Journal of Trend in Scientific Research and
Development. 2018, vol. 2(6), pp. 865–7.
[8] Dharanyadevi P., Therese M.J., Venkatalakshmi K. ‘Internet of things‐
based service discovery for the 5G‐VANET milieu’. 'Cloud and IoT–Based
Vehicular Ad Hoc Networks', Publisher: Wiley. 2021, pp. 31–45.
[9] Surbiryala J., Rong C. ‘Data recovery and security in cloud’. 9th International
Conference on Information, Intelligence, Systems and Applications (IISA),
Akynthos, Greece; IEEE, 2018.
[10] Therese M.J., Devi A., Kumar T. A. 'Interfacing FOG and cloud computing
for IOT applications'. Recent Developments In Computing, Electronics and
Mechanical Sciences. 2020.
[11] Raigonda M.A., Raigonda M.R., Raigonda M.R. 'A cloud based automatic
recovery and backup system with video compression'. International Journal
Of Engineering And Computer Science. 2016, vol. 5(9).
[12] Therese M.J., Ezhilarasi C., Harshitha K., Jayasri A. ‘Secured data partition
and transmission to cloud through FOG computing for IOT application’.
IJAST. 2020, vol. 29(11), pp. 921–31.
[13] Kumar S., Ranjan P., Ramaswami R. ‘EMEEDP: enhanced multi-­hop energy
efficient distributed protocol for heterogeneous wireless sensor network’.
Proceedings of the 5th International Conference on Communication Systems
and Network Technologies, CSNT 2015; Gwalior, India, 4-­6 Apr 2015;
2015–194–200
[14] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, IEEE, 2014.
[15] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India, 2021. pp. 1–6.
[16] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems; Singapore: Springer. THDC-­Institute of Hydropower
Engineering and Technology, Tehri, India, 2020.
[17] Kumar S., Trivedi M.C., Ranjan P. Evolution of Software-­Defined Networking
Foundations for IoT and 5g Mobile Networks. Hershey, PA: IGI Publisher;
2020. p. 350.
[18] Devi A., Julie Therese M., Premalatha G. ‘Cloud computing based intelli-
gent bank locker system’. Journal of Physics: Conference Series. 2021, vol.
1717(1), p. 012020.
 Cram on data recovery and backup cloud computing techniques  133

[19] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[20] Ali S.A., Affan M., Alam M. ‘A study of efficient energy management tech-
niques for cloud computing environment’. 9th International Conference on
Cloud Computing, Data Science & Engineering (Confluence); 2019.
[21] Hua Y., Liu X., Feng D. ‘Cost-­efficient remote backup services for enterprise
clouds’. IEEE Transactions on Industrial Informatics. 2016, vol. 12(5), pp.
1650–7.
[22] Yu Gu., Dongsheng Wang., Chuanyi Liu. ‘DR-­cloud: multi-­cloud based dis-
aster recovery service’. Tsinghua Science and Technology. 2014, vol. 19(1),
pp. 13–23.
[23] Chakraborty B., Chowdhury Y. ‘Disaster recovery: background’. Introducing
Disaster Recovery with Microsoft Azure. 2020, pp. 1–41.
[24] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[25] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
Personal Communications. 2021, vol. 10(3).
[26] Haidar M., Kumar S. Smart healthcare system for biomedical and health care
applications using aadhaar and blockchain. 5th International Conference on
Information Systems and Computer Networks (ISCON); Mathura, India, 22-­
23 Oct 2021; 2022. pp. 1–5.
[27] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic Torus
network-­on-­chip architecture’. International Journal of Innovative Technology
and Exploring Engineering. 2019, vol. 8(6), pp. 2278–3075.
[28] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[29] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[30] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring
using machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[31] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium, Jan; 2015. pp. 2363–7.
134  Intelligent network design driven by Big Data analytics

[32] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
Proceedings of the International Conference on Computational Intelligence
and Communication Networks, CICN 2015; 2016. pp. 79–84.
[33] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 4th International Conference on Information Systems and
Computer Networks, ISCON 2019; 2019. pp. 250–5.
[34] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of Indian roads’. Proceedings of the
Confluence 2020-­10th International Conference on Cloud Computing, Data
Science and Engineering; 2020. pp. 63–76.
[35] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and IoT for
smart road traffic management system’. Proceedings of the IEEE India
Council International Subsections Conference; INDISCON 2020; 2020. pp.
289–96.
[36] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation of
fault tolerance technique for Internet of things (IoT)’. Proceedings of the 12th
International Conference on Computational Intelligence and Communication
Networks, CICN 2020; 2020. pp. 154–9.
[37] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, Internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks; 2021. pp. 01–6.
 Chapter 7
An adaptive software-­defined networking (SDN)
for load balancing in cloud computing
Swati Lipsa 1, Ranjan Kumar Dash 1, and Korhan Cengiz 2

The Internet of Things can be perceived as a collection of millions of devices that

are connected among each other and with the internet as a connectivity backbone
to acquire and share real-­time data for providing intelligent services. The tremen-
dous rise in the number of devices requires an adequate network infrastructure to
remotely deal with data orchestration. To overcome this issue, a new approach of
infrastructure sharing over the cloud among service providers has transpired, with
the goal of lowering excessive infrastructure deployment costs. The software-­defined
networking (SDN) is a networking architecture that enables network operators and
users to monitor and manage the network devices remotely and more flexibly by
using software that runs on external servers. As SDN and cloud integration improves
reliability, scalability, and manageability, this chapter combines cloud infrastruc-
ture with SDN. Although SDN-­based cloud networks have numerous advantages
as mentioned above, there still exist certain challenges that draw the attention of
researchers like energy efficiency, security, load balancing, and so on. The work car-
ried out in this chapter is an attempt to address one of the challenging tasks, namely
the load balancing, by developing a new multiple-­controller load-­balancing strat-
egy. The proposed strategy effectively balances the load even if one or more super
controllers fail. Furthermore, results are simulated and compared under different
operational environments, both with and without the Modified Bully algorithm. The
comparison results ensure that the introduced technique exhibits better performance
with metrics such as packet loss, packet transmission ratio, and throughput.

7.1 Introduction

“The Internet of Things (IoT) is a network of physical objects (or ‘things’) equipped
with sensors, software, as well as other technologies that facilitate interaction and

1
Department of IT, Odisha University of Technology and Research, India
2
College of IT, University of Fujairah, UAE
136  Intelligent network design driven by Big Data analytics

exchange of information between devices and systems via the internet” [1]. This
means the Internet of Things (IoT) is a network that comprises internet-­enabled
smart devices that can gather, transmit, and process data from their application envi-
ronments using embedded devices like sensors, CPUs, and different types of com-
munication hardware. IoT devices can be connected to an IoT gateway to exchange
the acquired sensor data from various devices. The sensor data can then be pro-
cessed locally or routed to the cloud for further analysis. The data collected by these
devices can be translated into valuable information such as finding patterns, provid-
ing suggestions, helping in decision-­making, and identifying potential problems that
may arise without the need for human intervention in the majority of the cases. In
this manner, an IoT application interacts with smart devices that automate processes
to meet certain requirements. The application domain ranges from the medical and
healthcare sector to smart home, industrial automation, smart cities, smart grids,
transportation and logistics, retail, and so on. This paves the way for advanced ser-
vices aimed at raising the standard of living.
The IoT is perceived as a key frontier that has the ability to improve nearly all
aspects of our life. It enables humans to live and work smartly and achieve control
over their daily activities to a large extent. Many businesses have started to use this
technology because it gives firms a real-­time glimpse of how their systems operate,
providing insights into anything from machine performance to logistics operations
and supply chain management. This technology comes with numerous benefits,
some of which include: the capability to monitor and manage activities, saving
money, time, and resources by automating processes, allowing analytical decisions
to be made more quickly and precisely, improving the Quality of Service (QoS), and
enhancing customer experience. The IoT still has a lot of unexplored opportunities.
This technology may attain its full potential if all devices are able to interact with
one another, independent of brand or company.
As the number of devices appears to be escalating by leaps and bounds, and
there is an enormous amount of data being generated, the traditional network infra-
structure is struggling to catch up with this new digital era. The traditional network
infrastructure fails to effectively manage the network resources, especially during
peak hours. Traditional networking supports the old-­fashioned approach of network-
ing strategy that uses fixed and dedicated hardware devices like routers and switches
to monitor network traffic. Traditional networks are becoming more sophisticated as
more protocols are used to increase reliability and network speeds. The absence of
open and standard interfaces hinders interoperability. Network devices are limited in
their ability to evolve since they are built on proprietary hardware and software and
are static. The traditional networks were well adapted to static traffic patterns and
featured a plethora of protocols tailored to particular tasks. Due to the complexity of
traditional network components making modifications to networks to accommodate
flexible traffic patterns has grown extremely difficult.
Traditional networks have a high management overhead. In traditional networking
infrastructures, businesses are frequently tied to a specific vendor, without any standard
procedures for configuring devices across network providers. It is becoming increas-
ingly challenging for network administrators to manage various network devices and
 An adaptive software-­defined networking (SDN) for load balancing  137

interfaces from various vendors. A network upgrade necessitates alterations to the con-
figuration of numerous devices. Furthermore, networks must expand to accommodate
hundreds or thousands of new devices with varying performance and service require-
ments. When it comes to configuring the network system, the network provider faces
significant challenges in making modifications to the network system, as individual
devices must be manually programmed for any change in the network infrastructure.
Another issue is the degradation of scalability as well because networks are congested
and cannot be rapidly reconfigured to serve critical traffic. The conventional method
of network management is being rendered outdated by the increasingly large amounts
of data traffic, complicated network design, and growing expectations to enhance the
performance of the network. The inability to fulfill these demands is a big drawback of
traditional networks. These limitations make it difficult for IoT devices to communicate
in a versatile and reliable manner [2].
In the light of the aforementioned issues, Software-­Defined Networking (SDN)
is perceived to be a technology enabler for delivering adequate solutions [3]. It has
emerged as a modern and promising model for migrating from existing traditional
networks and facilitates users with more programmability and easier resource man-
agement. It is a networking architecture in which the control plane (network’s con-
trol functions) and data plane (packet forwarding) are decoupled and the network
controller is centralized. This architecture varies from conventional networks that
regulate network traffic through specialized hardware devices such as switches and
routers. SDN allows the use of software to establish and manage a virtual network,
as well as traditional hardware.
There are numerous motivating reasons underlying the development of SDN.
Some of them include: lead to faster provisioning and administration of network
resources, has the potential to simplify statically designed networks, the network can
be overviewed conveniently from a single place, offers cost-­effective operation as
administrative cost decreases and server utilization increases, adheres to open stan-
dards, and is compatible with network hardware from any manufacturer. The objec-
tive of SDN is to enable network administrators and enterprises to respond quickly
to changing business requirements. SDN can be updated swiftly and in bulk, with-
out reconfiguring each device manually. Since this architecture optimizes the net-
work operations and can be programmed through software, they are more versatile,
agile, and scalable as well as fit well with cloud architecture. As a result, SDN has
become an important component of a cloud-­based framework [4]. The advantages of
integrating SDN with cloud computing include operational cost reduction, security,
reduced downtime, cloud abstraction, etc. These capabilities of SDN-­enabled cloud
computing make this an ideal choice for effectively managing the dynamic nature
of IoT.
Despite its diverse capabilities and unique characteristics, SDN introduces new
challenges for network engineers and service providers emphasizing on energy effi-
ciency, performance, security, load balancing, and virtualization [5]. To make the
maximum utilization of an SDN-­based cloud network, the system must balance the
load among multiple devices in the network, and our work concentrates on harness-
ing the benefits of a balanced SDN network.
138  Intelligent network design driven by Big Data analytics

The rest of this chapter has been organized as follows: Section 7.2 delves into
related studies on load balancing with SDN controllers. Section 7.3 describes the
basic background related to SDN architecture in a subtle way. Section 7.4 specifi-
cally discusses the taxonomy of load balancing in SDN. Section 7.5 provides an
insight into the implementation of our proposed algorithms. An illustration of the
proposed algorithm is given in Section 7.6. The proposed mechanism’s performance
is discussed and the outcomes are analyzed in Section 7.7. The summary of the work
as well as future research potential is presented in Section 7.8.

7.2 Related works

The work carried out in [6] introduces a distributed architecture-­ based load-­
balancing technique for SDN controllers called a dynamic and adaptive algorithm,
which is demonstrated by implementing a prototype system based on floodlight. The
OpenFlow switch and controller packages [7] are used to create an SDN-­based cloud
computing platform that enables certain functions such as energy conservation, load
balancing, etc. To resolve the issue of managing elephant flows in data center net-
works, an SDN-­based load-­balancing (SBLB) strategy for elephants [8] is addressed
in this chapter that adapts multiple routing paths to distribute the load in response to
changing load conditions. The paper [9] presents OpenFlow Round-­Robin (RR) and
OpenFlow Least-­Connections algorithms to overcome the limitations of conven-
tional load-­balancing algorithms. It is found that the OpenFlow Least-­Connections
algorithm performs better as compared to the OpenFlow RR algorithm in balancing
the load and has a shorter response time. A middlebox based on the Clos network
and SDN is developed [10] to improve bandwidth usage while ensuring QoS for
data centers. The authors in [11] present a technique based on employing an appro-
priate configuration for the SDN controller operating parameters in order to mini-
mize the SDN controller computational load by modifying the activity of the control
functions with little influence on the efficiency of the functions. An architecture
facilitating software-­defined clouds is formulated in [5] that emphasizes on mobile,
web, and enterprise cloud applications, and the same is evaluated for two use cases,
namely, QoS-­aware bandwidth allocation and bandwidth-­aware energy-­efficient
VM (Virtual Machine) placement.
The dependence on a single controller can lead to scalability issues, so Sufiev
and Haddad [12] have proposed a multi-­controller architecture as a solution that
enables dynamic load balancing and reduces the interdependence between Super
Controller and Regular Controller. The load-­ balancing technique developed in
Reference [13] is based on multiple distributed controllers. Furthermore, the experi-
ment was done using floodlight and exhibits that this technique can dynamically bal-
ance each controller’s load while minimizing load-­balancing time. The paper [14]
presents an Ant Colony Optimization (LLBACO) for link load-­balancing approach
that can balance network link traffic, increase QoS, and reduce network overhead.
The work carried out in Reference [15] introduces an SDN-­enhanced Inter cloud
Manager (S-­ICM) to assign network flows in the crowded cloud network. The
 An adaptive software-­defined networking (SDN) for load balancing  139

comparison of S-­ICM with existing works RR allocation and honeybee foraging

algorithm (HFA) reveals that S-­ICM outperforms HFA and RR in terms of mitigat-
ing system saturation. Kanagavelu and Aung [16] implement an SBLB framework
with a load-­aware policy that is employed using an OpenFlow switch and a control-
ler. This approach distributes the load across the servers and can be used in lieu of
commercial load balancers.
The paper [17] outlines and categorizes load-­ balancing strategies in SDN
networks, as well as their benefits and limitations. The Nio technology [18] is
employed in place of the cluster method for developing a method for controller-­to-­
controller communication in multiple SDN controller architectures that effectively
balances the load between these (or the) multiple controllers. Greedy-­based Service
Orientation Algorithm and SA (Simulated Annealing)-­based Service Orientation
algorithm have been introduced in Reference [19] to minimize the data transmission
time of IoT devices and balance the load among various Service Functions (SF) by
forwarding fixed packets to the SF having the lowest load. A taxonomy is created
in Reference [20] to represent distinct features of SDN cloud computing that gives
an in-­depth description of each component, with an emphasis on data center power
management, traffic monitoring, network function virtualization, and security. To
enhance resource utilization and reduce user response time, an SBLB service for
cloud-­based servers is described in Reference [21]. A load adjustment approach is
used in Reference [22] to accomplish load balancing among multiple controllers by
applying the said approach to each controller.
This study [23] explores the load-­balancing mechanisms implemented in the
SDN, classifying them into two types: deterministic and nondeterministic. The per-
formance of centralized and distributed controllers is compared in Reference [24]
based on three algorithms, viz. random, RR, and weighted RR algorithm. The results
suggest that distributed controllers outperform centralized controllers in certain
parameters such as single point of failure, scalability, and response time, with the
weighted RR algorithm being the best among the three algorithms focused in this
chapter. The significance of load balancing on network performance metrics [25]
is visualized using three types of controllers, namely, hub-­like controller, spanning
tree protocol-­based controller, and a load-­balancing-­based controller. The compar-
ison aspects of formerly used controllers reveal that the network’s links are less
congested, which improves the overall performance of the network. The proposed
service orchestration and data aggregation methodology [26] collects data packets
and organizes data as services to minimize data redundancy and service response
time. The work in [27] adds to the discussion by categorizing current evolving load-­
balancing approaches in SDN as traditional and artificial intelligence-­based tech-
niques to improve QoS. A hierarchical edge-­cloud SDN controller system model is
built in [28] to optimize network scalability and scale down computation delay in
SDN networks under QoS criteria even while dealing with a large-­scale network.
From this cumulative study, we can have a view that the implementation of
SDN in a cloud computing environment is significantly essential in deploying net-
works that are customizable and manageable. The network controller in SDN is
centralized and facilitates automated network management. The controller acts as
140  Intelligent network design driven by Big Data analytics

the gateway between the user and the server. Due to the dense traffic inside the net-
work, issues of bottlenecks may arise at gateways. Hence, the load of such networks
discussed earlier needs to be distributed evenly to avoid congestion. While some
studies concentrated on static load balancing, some on dynamic load balancing, and
others on diverse multipath routing algorithms to control traffic, they all had one
thing in common: they were all about traffic management. There are also shreds of
evidence of developments that have integrated both load balancing and routing to
resolve the bottleneck at gateways.
But looking into the load-­balancing techniques in the previously explained algo-
rithms, they can be termed as practically insufficient to deliver a substantial amount
of efficiency even while working in the multi-­controller distributed SDN architec-
ture. In a distributed architecture, multiple controllers can be arranged either in a flat
or hierarchical manner. The controllers arranged in a hierarchical fashion typically
consist of a controller on top level that acts as the super controller and manages the
load among local controllers. However, the possibilities of a single point of failure
cannot be ruled out as there is only a single super controller in the hierarchy. The
existing studies pertaining to multiple controllers do not discuss the way of finding
an alternative in case of a super controller failure as mentioned earlier. Our proposed
algorithm promises to cater to the needs of an effective load-­balancing mechanism
by finding a replacement strategy in case of a super controller failure in a hierarchi-
cally distributed SDN architecture. This will prove to be an algorithm for niche
categories in the area of load balancing.

7.3 Architecture overview of SDN

A SDN architecture [29] specifies how a networking and computing infrastructure

can be designed using a combination of open and software-­based applications. SDN
eliminates the need for a tight coupling between network configuration and traf-
fic engineering from the underlying hardware infrastructure, enabling systematic
and consistent network control through open application programming interfaces
(APIs). This architecture is categorized into three distinct layers [30], as follows:

1. Infrastructure layer: networking devices that monitor the network’s routing and
data processing capabilities are orchestrated in this layer. These devices can be
a collection of switches and routers in a data center responsible for handling
packets according to the rules set by a controller. This layer deals with gathering
various network parameters like the flow of traffic, topology, network utiliza-
tion, etc., and sends them to the control layer. This layer acts as the physical
layer over which network virtualization is established using the control layer.
2. Control layer: this layer is perceived as the brain of SDN. The SDN controller
serves as a bridge between the application and infrastructure layer. This layer
owes its intelligence to the centralized SDN controller software that controls net-
work infrastructure. The controller accepts the application layer’s requirements
 An adaptive software-­defined networking (SDN) for load balancing  141

and conveys them to the networking devices. It also sends back the information
fetched from networking devices to the application layer.
3. Application layer: services and applications running in this layer describe the
behavior of a network. These applications are programs that employ APIs to
communicate their desired network requirements and behavior to the SDN con-
troller. Furthermore, these applications can provide the network’s abstract view
by obtaining information from the controller for appropriate decision-­making.
There are numerous types of applications that can be developed like network
configuration and management, intrusion detection systems, traffic monitoring
and control, business policies, etc. In the real world, these applications offer a
range of end-­to-­end solutions for corporate and data center networks.

The APIs act as the control point for all network components. The APIs in the
SDN framework are known as southbound and northbound interfaces that repre-
sent the communication between controllers, network devices, and applications. A
southbound interface enables a network component, i.e., controllers to interact with
lower-­level components, i.e., switches and routers. OpenFlow is a southbound API
used by the administrators to add, amend, and delete entries in the internal flow table
of network switches and routers allowing the network to accommodate real-­time
traffic demands. On the contrary, a northbound interface facilitates communication
between higher-­level components, i.e., it is the connection between the controller
and the application. The end-­user application conveys its requisites to the network,
such as data, storage space, bandwidth, and so on, and the network replies to the
application with the appropriate resource based on the resource availability.
Despite the fact that a controller utilizes just one-­third of the OpenFlow pro-
tocol, it is very vital. A controller manages and regulates all of the switches and
routers, creates a virtualized network, and forwards the incoming packets. Taking
into consideration the significance of controller in the SDN framework and the var-
iegation of architectures and deployments in the business and research fields, there
is a demand to evaluate and benchmark all of these options with respect to numerous
performance indicators. When the traffic increases due to the increase in dynamic
arrival of incoming packets, the potentiality of a single controller to provide service
for the overwhelming traffic decreases in terms of processing power, bandwidth lim-
its, and memory. This leads to the network relying on a single SDN controller facing
a catastrophic bottleneck making it the point of failure [31]. Due to this limitation,
the SDN network is susceptible to face issues related to packet loss, scalability, load
balancing, reliability, and performance degradation. Therefore, this situation neces-
sitates the role of multiple controllers to address concerns pertaining to load balanc-
ing and performance improvement.

7.4 Load-balancing framework in SDN

Load balancing refers to the approach of distributing computational tasks among a set
of resources. It is the technique of sharing network traffic across multiple servers. The
142  Intelligent network design driven by Big Data analytics

Figure 7.1   Network without load balancing

goal of load balancing is to deal with unexpected traffic surges, maximize throughput,
reduce latency, reduce response time, speed up the performance, and evade the over-
burden of any single resource. Figures 7.1 and 7.2 depict an overview of a network
without and with load balancing, respectively.

7.4.1 Classification of SDN controller architectures

Load balancing in the SDN framework may be classified into various subcategories
based on different viewpoints. This chapter classifies load balancing, as shown in
Figure 7.3.

7.4.1.1 Distributed architecture
Distributed architecture deals with the dispersion of loads over the entire set of oper-
ating nodes rather than relying completely on a single node. The controllers and
switches both act as nodes and have many-­to-­many interactions among themselves.
Here, every controller operates as a super controller, capable of making decisions
based on the network it is operating. However, before an overloaded controller can
decide to reroute incoming packets, it must first acquire load information from other
An adaptive software-­defined networking (SDN) for load balancing  143

Figure 7.2   Network with load balancing

Figure 7.3   Taxonomy of load balancing in SDN

144  Intelligent network design driven by Big Data analytics

controllers. This load information is acquired on a real-­time basis [32]. This approach
promises to exhibit improved scalability over centralized architecture. There are two
types of distributed control plane layouts, i.e., flat and hierarchical [11].

7.4.1.2 Hybrid architecture
Hybrid serves as an architecture where both distributed and centralized architecture
coexist [33]. This architecture has the advantage of distributed processing and cen-
tralized control where the distributed processing part is borrowed from distributed
architecture and the latter from the centralized architecture.

7.4.1.3 Centralized architecture
A specialized super controller is used in the centralized approach for gathering the load-
ing statuses of all other controllers and coordinating traffic management. This special-
ized super controller acts as a coordinator that maintains the global controller load infor-
mation table. When there is an uneven load distribution among controllers, the controller
decides to balance the load using the load information table. However, when the super
controller fails or becomes overloaded, the whole network comes down due to a single
point of failure, just like when only one controller is employed [22].

7.5 Problem statement

7.5.1 Selection strategy of controller head

Considering the global perspective of traffic density in the control plane, the super
controller incorporates a dispatcher strategy needed for controlling and monitoring
the load distribution of flow signals among the local controllers. The super control-
ler or controller head manages multiple local controller clusters, where each clus-
ter represents a network that consists of local controllers associated with multiple
number of switches and routers. Local controllers compute the routing path for the
incoming stream of signal packets acquired from both the super controllers and the
switches. Meanwhile, after computing the routing path, local controllers must com-
municate their own traffic conditions to their respective super controllers to ensure
that the super controller preserves the real-­time traffic flow information of all local
controllers.
As the super controller deals with the traffic administration of all local control-
lers, a single super controller due to its limited capacity might cause an unpleasant
delay during extreme traffic conditions. This reason brings forth the idea of having
multiple super controllers to efficiently balance the heavy traffic load among all
super controllers, i.e., if a controller fails abruptly, the system can be revived by
migrating all of the failed controller’s communication to other working controllers.
These multiple super controllers coupled together must be utilized in a coordinated
manner to provide load balancing. This coordinated fashion of controllers can be
achieved using the Bully algorithm in this scenario. “In distributed computing, the
Bully algorithm is a method for dynamically electing a coordinator or leader from
 An adaptive software-­defined networking (SDN) for load balancing  145

a group of distributed computer processes. The process with the highest process
ID number from among the non-­failed processes is selected as the coordinator”
[34,35,36,37,38,39,40,41,42].
There can be different strategies for the replacement of super controllers
depending on the kind of applications and the number of super controllers that fail or
become overwhelmed in certain instances of time. In case of failure of a single super
controller, the same can be replaced by another super controller, whereas the failure
of multiple super controllers can be handled by replacement of inactive super con-
trollers by another set of multiple super controllers. We have applied the Modified
Bully algorithm for efficient determination of such replacement strategies of super
controllers in case of failures.
The Modified Bully algorithm treats super controllers in the same way that the
Bully algorithm handles processes. A Modified Bully algorithm is a leader elec-
tion algorithm that dynamically elects a coordinator or leader from a collection of
passive super controllers. In this algorithm, weight is assigned to the passive super
controller based on the hop distance between the passive and active super controller.
The passive super controller with the highest assigned weight from among the avail-
able passive super controllers is selected as the candidate replacement for the failed
super controller [43,44,45,46,47,48,49,50,51].
When a super controller becomes overloaded and fails, the system undertakes
load balancing and redirects packet flows to another super controller with a lower
load. As a result, the proposed SDN architecture’s load management technique
focuses on coordination among all super controllers.
Modified Bully algorithm entails the following steps:

1. Each passive super controller has been assigned a unique weight with respect to
each active super controller.
2. Every active super controller knows the assigned weight to each of the passive
super controllers.
3. The passive super controller that detects the failure of the active super controller
initiates the election.
4. Several passive super controllers can initiate an election simultaneously, and
the passive super controller with the highest assigned weight is elected as the
new coordinator.

There are two types of super controllers, viz. active and passive. Each active
super controller exhibits two states, i.e., ACTIVE and FAILED, whereas each passive
super controller can take up two states, i.e., AVAILABLE and NOT_AVAILABLE.
146  Intelligent network design driven by Big Data analytics

‍ ‍

The function send(Si,Sj,‘ELECTION’) represents that Si sends ‘ELECTION’

message to Sj. Similarly, response(Sj) represents the response from Sj.

7.5.2 Network setup
Let S be a set of super controllers and Si be the ith super controller. Let N be the set
of networks and Nj be the jth network, such that for SiNj represents that the jth net-
work is managed by the ith super controller. A controller is active if it has at least
one network to handle, otherwise, it is passive. An active controller possesses a flow
table of the network under its control, whereas a passive controller maintains the
flow table of all networks controlled by the active controller. Let W be the weight
assigned to passive super controllers and Wk be the weight of kth super controller.
Let W’ be the weight assigned to active controllers such that W′ > W. Let m be the
number of packets to be processed by super controller Si and C be the capacity of
the super controller to process incoming packets such as C(Si) is the capacity of the
ith super controller.
 An adaptive software-­defined networking (SDN) for load balancing  147

‍ ‍

The function process(Si, m) represents that ith super controller processes m num-
ber of packets.

7.6 Illustration

The illustration of the above configuration (i.e Figure  7.4 and Figure  7.5) can be
explained as follows.
For instance, the S consists of seven Si, namely S1, S2, S3, S4, S5, S6, and S7, where
S1, S3, S5, and S7 are active and manage networks N1, N2, N3, and N4, respectively,
as shown in Figure 7.6. The super controllers S2, S4, and S6 are passive. Weights are
assigned to the passive controllers (i.e. S2, S4, and S6) on basis of the number of hops
from the starting point. Assuming that S1 is the starting point, S1 is one hop away
from S2, so weight assigned for S1 to S2= 4. Likewise, if only passive controllers are
included, the weight assigned for S1 to S6 = 3 and S1 to S4 = 2. In a typical scenario,
the following conditions might arise while working with such types of networking
environments.

i. S1 fail, S2 available:
Suppose S4 tries to communicate with S1 and receives no reply from S1.
So S4 gets to know that S1 has failed and starts the election for replacing
148  Intelligent network design driven by Big Data analytics

Figure 7.4   Network configuration of the super controller and local controller

S1. To start with S4 broadcasts ELECTION messages to passive super

controllers assigned with higher weight than S4, i.e., S6, S2, and S1.
Similarly, S6 also broadcasts ELECTION messages to passive super con-
trollers assigned with higher weight than S6, i.e. S2, and S1, and S2 being
the passive super controller with the highest weight sends ELECTION
messages only to S1. Both S6 and S2 reply to S4 with OK messages to
confirm that they will participate in the election, whereas S1 sends no
reply owing to its failed state. Similarly, S2 also replies to S6 with an OK
message to confirm that it will participate in the election and again S1
sends no reply. As S4 and S6 have received OK messages from S2 who is
higher in weight from both of them, so now S4 and S6 are dropped from
the election process and wait for COORDINATOR message as other
passive super controllers with higher weight are actively participating
in the election. As S2 is the passive super controller with the highest
weight and never receives an OK message from any other passive super
controller, it is now elected as the replacement for the failed S1 and S2
sends out COORDINATOR messages to the rest of the passive super

Figure 7.5   Super controller implementation framework

 An adaptive software-­defined networking (SDN) for load balancing  149

Figure 7.6  Illustration of super controller consisting of multiple super

controllers

controllers, i.e., S4 and S6. It does not send a COORDINATOR message

to S1 as it is failed.

ii. S1 fails, S2 not available, and S6 available:

If S1 fails then S2 is the most preferable replacement for S1 because of
its highest weight. But if S2 is not available (i.e. failed or overloaded),
then the same election process as described for the first condition will
be initiated and S6 being the next highest weight passive super controller
after S2 will be elected as the replacement for the failed S1.

iii. S1 fail, substitute selected and higher weighted Si revives or available:

In this condition, S1 has failed and S2 is selected as the replacement for
the failed S1 in the same procedure as discussed for the first condition.
But whenever S1 revives after S2 is selected for the replacement, S1 being
the highest weighted super controller does not start an election rather
it sends a COORDINATOR message to all passive super controllers
including S2, which implies it is available and can handle its own packet
and S2 is set free.
150  Intelligent network design driven by Big Data analytics

Figure 7.7   Runtime vs. throughput

7.7 Results and discussion

The simulation of the proposed model is performed by using the python program-
ming language with the help of the POX tool. POX tool is an open-­source platform
for SDN controllers like OpenFlow. The super controllers as well as the switches are
configured to operate over the virtual networks.
The following two criteria are taken into consideration for the simulation of the
proposed model:
Case I: Allowing the super controller to transmit packets to the appropriate net-
work without using the Modified Bully algorithm, even if one or more super control-
lers are down.
Case II: When a super controller fails, the Modified Bully algorithm is employed
to find a replacement for it.
Both scenarios are compared to determine the performance in terms of various
metrics such as throughput, packet transmission ratio (PTR), and packet loss (as
shown in Figures 7.7–7.9, ).

7.7.1 Comparison of throughput
The observations of the simulation of the two scenarios are recorded once one or
more super controllers start failing (i.e. after 100 s of operation), the reason being
 An adaptive software-­defined networking (SDN) for load balancing  151

Figure 7.8   Number of packets per second vs. packet transmission ratio

they behave similarly in the absence of failure (Figure  7.7). This figure depicts
the overall network throughput over runtime with a moderate number of packets
per second. Both of the aforementioned situations have a nearly identical perfor-
mance by fluctuating over two levels of throughput. However, implementation of
the Modified Bully algorithm for load balancing results in a remarkable increase
in overall system throughput compared to not using the Modified Bully algorithm,
because the Modified Bully algorithm easily finds a replacement to combat super
controller failure.

7.7.2 Comparison of PTR
The PTR is calculated using the following formula:
Ps
PTR = (7.1)
‍ pT ‍
where ‍PS ‍ = number of packets processed successfully and ‍PT ‍ = total number of
packets received.
By using (7.1), the PTR is evaluated for both cases, as shown in Figure 7.8. This
figure indicates how the PTR of the total network increases when the number of
packets received per second increases up to a certain value beyond which the PTR of
the system shows a gradual decrease with further increase in the number of packets
152  Intelligent network design driven by Big Data analytics

Figure 7.9   Number of packets per second vs. number of packet loss

received per second, due to increase number of packet loss. Here these values are
termed as threshold values which are 0.92 at 120 packets/s and 0.98 at 175 packets/s
for the two cases, respectively.

7.7.3 Comparison of number of packet loss

Figure 7.9 demonstrates the amount of packet loss as a percentage of the total num-
ber of packets generated per second. The increase in the number of packets increases
the quantity of data to be transferred; this raises the likelihood of packet errors. It is
observed that the loss of packets in our proposed algorithm is considerably less than
the scenario where there is no implementation of the Modified Bully algorithm. The
application of the Modified Bully algorithm will have a significant impact on reduc-
ing the amount of packet loss due to lesser congestion inside the network.

7.8 Conclusion

In spite of the features of SDN that emerge as a break-­through technique for han-
dling complicated networking operational difficulties, there are some possibilities
of a computational resource-­limited controller being overwhelmed by heavy traffic
and then experiencing unexpected delays. A multi-­controller architecture evolves as
a solution to the issues faced by the single controller, but uneven load distribution
 An adaptive software-­defined networking (SDN) for load balancing  153

in multi-­controller deployment systems remains an inevitable concern. To avoid

this unbalanced load congestion and escalate network performance, we have intro-
duced a hierarchical multiple-­controller load-­balancing algorithm. Our proposed
algorithm dynamically manages network flow by distributing the workload across
multiple super controllers to optimize network resources. The result indicates that
our approach enhances the network efficiency and flexibility, and boosts the overall
network performance, by minimizing the overload and the delay between control-
lers. We aspire to analyze more parameters in the future and deploy this interface in
a real testbed to demonstrate its efficacy in real-­world scenarios.

References

[1] Aazam M., Khan I., Abdullah Alsaffar A., Huh E.-N. ‘Cloud of things:
Integrating Internet of things and cloud computing and the issues involved’.
11th International Bhurban Conference on Applied Sciences & Technology
(IBCAST); Islamabad, Pakistan, 14-­18 Jan; 2014.
[2] 'Traditional network infrastructure model and problems as-
sociated with it'. Journal of Pluribus Networks. 2012, vol.
14(2). Available from https://pluribusnetworks.com/blog/​
traditional-network-infrastructure-model-and-problems-associated-with-it/
[3] Ojo M., Adami D., Giordano S. ‘A SDN-­iot architecture with NFV imple-
mentation’. IEEE Globecom Workshops (GC WKSHPS); Washington, DC,
USA, IEEE, 4-­8 Dec, 2016. pp. 1–6.
[4] Barros S. ‘Applying software-­defined networks to cloud computing’. 33rd
Brazilian Symposium on Computer Networks and Distributed Systems;
Vitória, ES, Brazil, 18-­22 May; 2015.
[5] Buyya R., Calheiros R.N., Son J., Dastjerdi A.V., Yoon Y. ‘Software-­defined
cloud computing: architectural elements and open challenges’. International
Conference on Advances in Computing, Communications and Informatics
(ICACCI); Delhi, India, 24-­27 Sep; 2014.
[6] Zhou Y., Zhu M., Xiao L., et al. ‘A load balancing strategy of SDN controller
based on distributed decision’. IEEE 13th International Conference on Trust,
Security and Privacy in Computing and Communications; Beijing, China, 24-­
26 Sep; 2014.
[7] Yen T.-C., Su C.-S. ‘An SDN-­based cloud computing architecture and its
mathematical model’. International Conference on Information Science,
Electronics and Electrical Engineering; Sapporo, Japan, 26-­28 Apr; 2014.
[8] Liu J., Li J., Shou G., Hu Y., Guo Z., Dai W. ‘SDN based load balancing mech-
anism for elephant flow in data center networks’. International Symposium on
Wireless Personal Multimedia Communications (WPMC); Sydney, NSW, 7-­
10 Sep; 2014.
[9] Zhang H., Guo X. ‘SDN-­based load balancing strategy for server cluster’. IEEE
3rd International Conference on Cloud Computing and Intelligence Systems;
Shenzhen, 27-­29 Nov; 2014.
154  Intelligent network design driven by Big Data analytics

[10] Tu R., Wang X., Zhao J., Yang Y., Shi L., Wolf T. ‘Design of a load-­balancing
middlebox based on SDN for data centers’. IEEE Conference on Computer
Communications Workshops (INFOCOM WKSHPS); Hong Kong, China, 26
Apr-­1 May; 2015.
[11] Caba C., Soler J. ‘Mitigating SDN controller performance bottlenecks’.
24th International Conference on Computer Communication and Networks
(ICCCN); Las Vegas, NV, 3-­6 Aug; 2015.
[12] Sufiev H., Haddad Y. ‘A dynamic load balancing architecture for SDN’. IEEE
International Conference on the Science of Electrical Engineering (ICSEE);
Eilat, Israel, 16-­18 Nov; 2016.
[13] Yu J., Wang Y., Pei K., Zhang S., Li J. ‘A load balancing mechanism for mul-
tiple SDN controllers based on load informing strategy’. 18th Asia-­Pacific
Network Operations and Management Symposium (APNOMS). IEEE;
Kanazawa, Japan, 5-­7 Oct; 2016.
[14] Wang C., Zhang G., Xu H., Chen H. ‘An ACO-­based link load-­balancing al-
gorithm in SDN’. 7th International Conference on Cloud Computing and Big
Data (CCBD); Macau, China, 16-­18 Nov; 2016.
[15] Kang B., Choo H. ‘An SDN-­enhanced load-­balancing technique in the cloud
system’. The Journal of Supercomputing. 2018, vol. 74(11), pp. 5706–29.
[16] Kanagavelu R., Aung K. ‘Software-­defined load balancer in cloud data cent-
ers’. Proceedings of the 2nd International Conference on Communication and
Information Processing; Singapore, 26-­29 Nov; 2016.
[17] Li L., Xu Q. ‘Load balancing researches in SDN: a survey’. 7th IEEE
International Conference on Electronics Information and Emergency
Communication (ICEIEC); Macau, China, 21-­23 Jul; 2017.
[18] Benamrane F., Mamoun M.B., Benaini R, Ben Mamoun M., Redouane B.
‘New method for controller-­ to-­
controller communication in distributed
SDN architecture’. International Journal of Communication Networks and
Distributed Systems. 2017, vol. 19(3), pp. 357–67.
[19] Chien W.-C., Lai C.-F., Cho H.-H., Chao H.-C. ‘Journal of Network and
Computer Applications’. A SDN-­SFC-­based service-­oriented load balancing
for the IoT applications. 2018, vol. 114, pp. 88–97.
[20] Jungmin S., Buyya R. ‘A taxonomy of software-­defined networking (SDN)-­
enabled cloud computing’. ACM Computing Surveys. 2018, pp. 1–36.
[21] Abdelltif A.A., Ahmed E., Fong A.T., Gani A., Imran M. ‘SDN-­based load
balancing service for cloud servers’. IEEE Communications Magazine. 2018,
vol. 56(8), pp. 106–11.
[22] Wang K.-Y., Kao S.-J., Kao M.-T. ‘An efficient load adjustment for balancing
multiple controllers in reliable SDN systems’. IEEE International Conference
on Applied System Invention (ICASI), Chiba, Japan; 2018.
[23] Neghabi A.A., Jafari Navimipour N., Hosseinzadeh M., Rezaee A. ‘Load
balancing mechanisms in the software defined networks: a systematic and
comprehensive review of the literature’. IEEE Access. 2018, vol. 6, pp.
14159–78.
 An adaptive software-­defined networking (SDN) for load balancing  155

[24] Ahmed H., Ramalakshmi R. ‘Performance analysis of centralized and dis-

tributed SDN controllers for load balancing application’. 2nd International
Conference on Trends in Electronics and Informatics (ICOEI); Tirunelveli,
India, 11-­12 May; 2018.
[25] Chahlaoui F., El-­Fenni M.R., Dahmouni H. ‘Performance analysis of load bal-
ancing mechanisms in SDN networks’. Proceedings of the 2nd International
Conference on Networking, Information Systems & Security; Rabat Morocco,
27-­29 Mar; 2019.
[26] Liu Y., Zeng Z., Liu X., Zhu X., Bhuiyan M.Z.A, Alam Bhuiyan M.D. ‘A
novel load balancing and low response delay framework for edge-­cloud net-
work based on SDN’. IEEE Internet of Things Journal. 2019, vol. 7(7), pp.
5922–33.
[27] Belgaum M.R., Musa S., Alam M.M., Su'ud M.M. ‘A systematic review of
load balancing techniques in software-­defined networking’. IEEE Access.
2020, vol. 8, pp. 98612–36.
[28] Lin F.P.-C., Tsai Z. ‘Hierarchical edge-­cloud SDN controller system with op-
timal adaptive resource allocation for load-­balancing’. IEEE Systems Journal.
2019, pp. 265–76.
[29] Haleplidis E. ‘Overview of RFC7426: SDN layers and architecture terminol-
ogy’. IEEE Softwareization; IEEE, 2017.
[30] Rana D.S., Dhondiyal S.A., Chamoli S.K. ‘Software defined networking
(SDN) challenges, issues and solution’. International Journal of Computer
Sciences and Engineering. 2019, vol. 7(1), pp. 884–9.
[31] Salman O., Elhajj I.H., Kayssi A., Chehab A. ‘SDN controllers: A compara-
tive study’. 18th Mediterranean Electrotechnical Conference (MELECON);
Lemesos, Cyprus, 18-­20 Apr; 2016.
[32] Ma Y.-W., Chen J.L., Tsai Y.H., Cheng K.H., Hung W.C. ‘Load-­balancing
multiple controllers mechanism for software-­defined networking’. Wireless
Personal Communications. 94; 2017. pp. 3549–74.
[33] Amin R., Reisslein M., Shah N. ‘Hybrid SDN networks: a survey of existing
approaches’. IEEE Communications Surveys & Tutorials. 2018, vol. 20(4),
pp. 3259–306.
[34] Haidar M., Kumar S. ‘Smart healthcare system for biomedical and health care
applications using aadhaar and blockchain’. 5th International Conference
on Information Systems and Computer Networks (ISCON); Mathura, India,
IEEE, 22-­23 Oct 2021, 2022. pp. 1–5.
[35] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
Personal Communication. 10; 2021. pp. 1–19.
[36] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
156  Intelligent network design driven by Big Data analytics

[37] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[38] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic Torus
Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering. 2019, vol. 8(6), pp. 2278–3075.
[39] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[40] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[41] Singh P., Bansal A., Kamal A.E., Kumar S. 'Road surface quality monitoring
using machine learning algorithm' in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent manufacturing and energy sustain-
ability. smart innovation, systems and technologies. Vol. 265. Singapore:
Springer; 2022.
[42] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and IoT for
smart road traffic management system’. IEEE India Council International
Subsections Conference (INDISCON); Visakhapatnam, India, 3-­4 Oct; 2020.
pp. 289–96.
[43] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementa-
tion of fault tolerance technique for internet of things (IoT)’. Proceedings
of the 12th International Conference on Computational Intelligence and
Communication Networks (CICN); Bhimtal, India, 25-­26 Sep; 2020. pp.
154–9.
[44] Singh P., Bansal A., Kumar S. ‘Performance analysis of various informa-
tion platforms for recognizing the quality of indian roads’. Proceedings of
the 10th Confluence of the International Conference on Cloud Computing,
Data Science and Engineering; Noida, India, IEEE, 29-­31 Jan, 2020. pp.
63–76.
[45] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India, 22-­23 Oct; 2021. pp. 1–6.
[46] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 4th International Conference on Information Systems and
Computer Networks, ISCON; Mathura, India, IEEE, 21-­22 Nov, 2019. pp.
250–55.
[47] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’. 2015
International Conference on Computational Intelligence and Communication
Networks, CICN; Jabalpur, India, IEEE, 12-­14 Dec, 2016. pp. 79–84.
 An adaptive software-­defined networking (SDN) for load balancing  157

[48] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium; Prague; 2015. pp. 2363–7.
[49] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wireless
sensor network’. 5th International Conference on Communication Systems
and Network Technologies, CSNT; Gwalior, India, IEEE, 4-­6 Apr, 2015. pp.
194–200.
[50] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, IEEE, 7-­8 Feb, 2014.
[51] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based trans-
parent and secure decentralized algorithm’. International Conference on
Intelligent Computing and Smart Communication 2019; Singapore; 2020.
This page intentionally left blank
 Chapter 8
Emerging security challenges in cloud
computing: an insight
Gaurav Aggarwal 1, kavita Jhajharia 2, Dinesh Kumar
Saini 2, and Mehak Khurana 3

Cloud computing has been evolved as a new computing prototype with the aim of
providing reliability, quality of service and cost-­effective with no location barrier.
More massive databases and applications are relocated to an immense centralized
data center known as the cloud. Cloud computing has enormous benefits of no need
to purchase physical space from a separate vendor instead of using the cloud, but
these benefits have security threats. The resource virtualization, the data and the
machine are physically absent in the cloud; the storage of data in the cloud causes
security issues. An unauthorized person can penetrate through the cloud security
and can cause data manipulation, data loss or theft might take place. This chapter
has described cloud computing and various security issues and challenges that are
present in different cloud models and cloud environment. It also gives an idea of
different threat management techniques available to encounter security issues and
challenges. The RSA algorithm implementation has been described in detail, and
the Advance Encryption Standard policy, along with its implementation, has also
been discussed. For better clarification, several reviews are conducted on the exist-
ing models.

8.1 Introduction

8.1.1  An introduction to cloud computing and its security

The unique term ‘cloud’ [1, 2] was taken from the diagrams of a computer network,
where the diagram hides the critical complexity of infrastructure associated within
a mesh of computer network. Cloud computing technology provides many services
including software as a service (SaaS), platform as a service (PaaS) and infrastructure

1
Department of Information Technology and Engineering, Amity University in Tashkent, Uzbekistan
2
Scool of Computing and Information Technology, Manipal University Jaipur, Jaipur, India
3
Department of Computer Science and Engineering, The NorthCap University, Gurugram, India
160  Intelligent network design driven by Big Data analytics

as a service (IaaS) that are the primary concern in cloud computing technology. The
main aspects of cloud computing technology consist of sharing of resources, services
on-­demand, compatibility of connectivity with various devices, perfect accountabil-
ity of services and complete access of network. The cloud reduces the investment
cost and software license cost as well. The main concern remains with the cloud is
the security of valuable data and its computing. The cloud storage services allow
data owner to move valuable data from the local machine to remotely located cloud
storages, which release cloud user’s impediment for managing as well as maintain-
ing their data at local devices. Cloud computing is a distributed architecture, and
its objective is clear. It is dedicated to providing convenient data storage and quick
computing through virtualizations of computing resources. It enhances the collabo-
ration, scalability, flexibility to adapt rapidly changing new technology and provide
overall cost reduction through efficient computing. Cloud computing continues to
evolve in the new era of internet-­based technology with reliable, ubiquitous and on-­
demand services. Cloud computing is a combination of service-­oriented architecture
[3], virtualization and advanced technologies with a strong dependency on internet-­
based applications. Google has introduced the MapReduce [2, 4] framework along
with Apache’s Hadoop Distributed File System (HDFS) [5], which is processing a
large amount of shared data with less time using the internet-­based applications.
In the present days, the enterprises are combining private and public cloud serv-
ers due to virtualization, and global storages of data and hence security for storage
of data and computation stole the spotlight. The storage of containers at the server
and moving them between multiple cloud environments need greater security in data
at rest and as well as in moving stages. In the present era, uses of cloud have been
increased at a larger scale.
There are online transactions, customer services, email accessing, online exams,
banking services, e-­governance, and so on. All these services are run on a cloud-­
based platform. So the security of those services requires enormous attention. Cloud
computing is dependent on vulnerable internet. The involvement of heterogeneous
architectures with different Cloud Service Providers (CSPs) raises more significant
security threats (Figure 8.1). Moreover, CSPs store the user’s data at different loca-
tions without knowing customers that also invite security issues. So, the security
paradigm needs to be changed to a new dimension.
The so-­called security concepts such as authorization, authentication and identi-
fication are no longer stand firm to prevent valuable data stored in the server of CSPs
by a cloud user and its safe transaction within the network from possible threats.
Several essential management techniques with encryption and decryption strategies
are adapted to safeguard the user’s data when it is stored in the cloud storage and on
transmission in the network as well. The data are encoded before it is warehoused
in cloud storage.
Numerous techniques are adopted to protect the data when it is traveling through
the network’s nodes. Despite undertaking so many precautions, security remains a
big question.
Are the data safe? Are the security measures taken so far appropriate? Due to
rapid changes in technology, security architectures also need further enhancement.
 Emerging security challenges in cloud computing: an insight  161

Figure 8.1    Encryption policy

It is well known that whenever the data are traveling in the insecure network, there
is no such fool-­proof technique which can assure complete security for user’s data.

8.2 The security issues in different cloud models and threat

management techniques

Before we move toward the details of data security and its transaction in the cloud
environment, we need to know about the characteristics of cloud architecture.
According to the definition of the National Institute of Standards and Technology
(NIST) [6], a complete model of the cloud is built up with five most important
features, in which three are the service models and four deployment models. Every
cloud model has different security issues, which need to be addressed in detail.

8.2.1  Five most indispensable characteristics

A cloud model should always have the below-­mentioned five essential characteris-
tics [6, 7] for efficient acceptance of the cloud users.

1. Broad network access: The cloud model should have open standards for all
Applications Running Interfaces (APIs). It should be able to function on all
IP.HTTPs and REST protocols and resources should be available to users from
anywhere with an internet connection.
2. Rapid elasticity: Cloud model should always be able to allocate the resources
among the users dynamically. It should release the additional resources
162  Intelligent network design driven by Big Data analytics

dynamically to the users as and when needed. The cloud model should be fully
automated.
3. Measured service: The cloud services are metered, like a utility, and the users
only pay for services used by them. The most important is that the services can
be canceled at any time by the users.
4. On-­demand self-­service: In the cloud model, users are abstracted from the
implementation, and the service is based on time (real-­time) delivery. The ser-
vices are accessed through a self-­service network interface.
5. Resource pooling: The requested resources are pooled from the common
sources, which builds a scale of economies. Common infrastructure gains high
efficiency and runs at greater network bandwidth.

8.2.2   The security issues in cloud service model

A cloud service model provides three types of services to the users as follows [2, 8]:

1. SaaS
2. PaaS
3. IaaS

Figure 8.2    Software as a service model

 Emerging security challenges in cloud computing: an insight  163

Figure 8.3    Platform as a service model

8.2.2.1 Security issues in software as a service model

In the security issues in software as a service (SaaS) model (Figure 8.2) , the CSP
provides various applications running in cloud infrastructure, which are available
as a service to the customer. The various applications are available and accessible
to users through a web-­based device like an email (web browser). In this service,
required software, operating system and network are provided by SaaS vendors [9].
The SaaS users have no control over the data security because all managements are
done by SaaS providers only. As the applications are provided through an open net-
work, any flaws in web applications and attackers may violet the data security at any
point of time. Traditional security protections are not strong enough in this scenario
for the adequate protection of data from possible threats. The customers have their
instances of applications which may be customized as per their needs.
Each customer gets different instances from the vendor and providers depend
on multitenancy model which create significant security issues in the cloud environ-
ment. The backup from third party providers also raises many security issues like
confidentiality, vulnerability, etc.

8.2.2.2 Security issues in PaaS model

In PaaS model (Figure  8.3), consumers are encouraged to deploy their software
or acquired applications to the cloud platform provided by the vendors. Users are
permitted to use various computer software languages, available libraries, other
164  Intelligent network design driven by Big Data analytics

Figure 8.4    Infrastructure as a service model

services and tools offered by the service providers. The network servers, operating
systems or storages are managed by the providers only. The security of PaaS ser-
vices depends upon protected and dependable networking and a safe web browser.
In a PaaS platform, there are two types of securities that are needed to be maintained
by service providers. The first one is the security and safety of the PaaS stage itself,
i.e., the runtime machine and another one is the security of the user’s applications
deployed on the PaaS platform. The PaaS owner generally uses third-­party service
components which are called as mashups [10, 11], a combination of many source
elements into a single combined unit. Mashups are more prone to threats when they
are not handled carefully on open networks.

8.2.2.3 Security issues in IaaS model

In IaaS (Figure 8.4), the consumers are capable of installing and running their soft-
ware or applications and related operating systems for processing, storing, network-
ing and other fundamental resources. The customers need not manage the cloud
infrastructures but have the control to manage over their storage system, operat-
ing systems and installed applications. In the IaaS model, service providers offer a
pool of resources like servers, storages and other infrastructures through virtualiza-
tion. In virtualization systems, consumers are using the same machine for running
different applications at a time, and they are also allowed to generate, duplicate,
modify, share, gist and push back virtual machines. The complexity of interconnec-
tions among various virtual machines and user’s applications leads to many security
 Emerging security challenges in cloud computing: an insight  165

Figure 8.5    The complete cloud model

threats. Though Virtual Machine Monitor (VMM) [12, 13] software is used for the
separation of virtual machines which are running simultaneously; flaws in VMM
applications need to be sifted through because any single snag in VMMs can break
down the whole systems.

8.2.3  Security issues in cloud deployment models

A cloud deployment model can have four different categories as follows:

1. Private cloud
2. Public cloud
3. Community cloud
4. Hybrid cloud

In these types of cloud models (Figure 8.5), the service providers provide the
complete package of networking, data storage, platform and software infrastructures
to the cloud users.

8.2.3.1 Security issues in private cloud

When cloud infrastructures are deployed for exclusive use for a single organization
having manifold users, this is a business model where it is possessed, administered
and operated by a single organization or a third party or may be managed by the col-
laboration of both. In the private cloud, resources and applications which are made
available by the cloud providers are pooled together for sharing the resources by the
cloud users. In view of security issues, though the private cloud is more secure than
any other deployment model, the authenticity of the users is needed to be guaranteed
166  Intelligent network design driven by Big Data analytics

carefully. Any unauthorized use may break the security wall and create a threat to
the cloud. In a private cloud, CSPs need to maintain and resolve all security issues.
The main advantage of the private cloud is that all resources are shared among the
authorized private personnel and bound to internal use only.

8.2.3.2 Security issues in public cloud

When cloud infrastructures are set up for open use for the general public, it is han-
dled and operated by some government or business organizations. It exists in the
premises of cloud providers. In the public cloud, all the resources are common
and shared among the customers dynamically, and a computing mechanism which
is called pay-­per-­use utility is maintained. This type of cloud is less secure as the
resources are shared among the public on-­demand basis. It is vulnerable to mali-
cious attacks by a third party. Robust authentication techniques are needed to ensure
secure user data. Examples of such types of cloud are Microsoft Azure and Google
App Engine.

8.2.3.3 Security issues in community cloud

Open Cirrus Cloud Computing Testbed, a collection of federal data centers across
six sites stretching from North America to Asia, is a typical example of community
cloud, which is when a cloud infrastructure is shared by numerous organizations in
a community and is managed by a third party. Another example of community cloud
is Facebook which is most popular nowadays. The community cloud model is less
secure in comparison to other deployment models. There is no robust mechanism to
separate users from each other in the cloud deployment. Account hijacking, stealing
of information and privacy of data are affected in this type of model. So the security
needs to be strengthened enough to safeguard user’s data.

8.2.3.4 Security issues in hybrid cloud

It is a hybrid of private and public clouds, in which a private cloud is connected to
one or more external services and functions as a single unit within a secure network.
It has an open architecture that allows it to connect to other management or super-
visory systems. Amazon Web Service is an example of a hybrid cloud (AWS). The
security of data storage is provided through applications acting as a firewall, spam
filter, etc. Because of the collaboration of different models, it is more prone to secu-
rity threats. There may be issues like Service-­Level Agreement (SLA) [14] among
the cloud providers (SaaS, PaaS and IaaS providers) and cloud users. The service
providers may blame each other for any shortcomings arise upon the security issues
of the applications, servers or tools.

8.2.4  Security challenges in the cloud environment

The primary goal of cloud computing is to increase processing power such that
millions of instructions can be executed each second (Figure 8.6). Cloud comput-
ing is made up of two parts: the user end and the cloud end. The user end includes
 Emerging security challenges in cloud computing: an insight  167

Figure 8.6    The cloud computing environment

the user’s devices, such as a computer, laptop, mobile phone, or any other access
device, as well as any application software that requires cloud services access. The
cloud end consists of high-­speed computing devices, servers and distributed data-
base system which provide the required access to the clients and hence form a cloud
environment.
The user end needs to connect his accessing devices like PC, laptop or mobile
to the cloud to access his data stored in a server in the cloud through an interface
software using the internet. The cloud being a distributed architecture, the user data
are stored in CSP’s server, and the place of the server is unidentified to the clients
and the users.
The CSP’s servers are managed by administrative groups to whom users need
to depend upon for safeguarding data and maintaining its privacy. The administra-
tive group members, who are trading with user’s data, can be a threat to secrecy and
privacy of user’s data. A threat management policy ensures that the cloud should
not be able to determine any evidence about the client’s data. The followings are the
security challenges which need immense attention.

8.2.4.1 Vulnerabilities
Applications like Gmail, Yahoo or Facebook are provided to users via an internet
browser. Attackers are penetrating the client’s computer or application by using the
web browser. Customary security solutions are not sufficient to protect data from
attacks. Consequently, advanced measures are required to be enforced. In the virtu-
alization technique, varied instances running on the same corporeal machine need
to be secluded. A VMM [15, 16] software is implemented to separate substantial
resources from different virtual machines running simultaneously. There are vul-
nerabilities in Microsoft Virtual PCs or Microsoft Virtual Servers that can allow a
casual visitor or third party to run malicious code on the host or another operating
system. Two virtual machines using a covert channel can communicate with each
other bypassing all the rules defined by the VMMs. Another cloud threat is called
168  Intelligent network design driven by Big Data analytics

Threat 11 [17]. In this, an attacker creates a negative Virtual Machine image con-
taining malware or virus. Then the attackers publish it on the provider’s storage area.
The users then retrieve them and infect the whole cloud environment.
To counter this attack, Mirage [18], an image management system, was pro-
posed, which focuses on the access control mechanism, image filtrations, derivation
tracking system and warehouse maintenance services. VMs are to be root-­locked, so
that virtualized guest environment is not permitted to interface with the host systems.

8.2.4.2 Attack in networks
8.2.4.2.1  Sniffer attacks
Data in a network travel from one node to another as a packet. Capturing of these
packets by intruders in a network is termed as sniffer attack. The data packets which
are not encrypted may be accessed and modified by the unauthorized guests, and
hence, vital data packets may lose its identity. A sniffer detection program through
the Network Interface Card guarantees that the data linked to specific systems in the
network are recorded so that its identity of data remains intact. Address Resolution
Protocol (ARP) [19] is used to detect sniffing attacks, where it maps the IP address
with the MAC address of the machine so that the data travel toward the designated
machine only. Another method called Round Trip Time [19] is also used in a sniffing
detection platform to detect a sniffing attack in networks. Here also attackers try to
copy the MAC address and capture the data by using their own modified software.

8.2.4.2.2  Spoofing attack

When a malicious party launches attacks which target network hosts and steal data,
it then spreads malware or bypasses access controls by imitating another user’s
device. Internet Protocol (IP) address spoofing, ARP spoofing, and Domain Name
Server (DNS) server spoofing are all examples of typical spoofing attacks. To attack
a network, attacker must link his MAC address and IP address of any actual existing
user in the network. The attacker sends spoofed IP address. Here the data which is
meant for the original user are sent to the attacker, and the attacker steals the infor-
mation and modifies data; this type of data is also included in session hijacking and
man-­in-­the-­middle attacks.
A DNS associates an IP address to a domain name and vice versa. DNS resolves
Uniform Resource Locators (URLs), email IDs and other readable domain names in
their respective IP addresses. In this type of attack, the attackers modify the server
to redirect the domain name to a different IP address and infect the data with mal-
ware. Spoofing attacks can be mitigated to some extent by utilizing packet filtering,
spoofing detection software and cryptographic network protocols (TLS, SSH and
HTTPs).

8.2.4.3 Reused IP address
This is a big security issue in terms of network security. Users migrate from one
network to another network frequently. So the old user when migrated from the
 Emerging security challenges in cloud computing: an insight  169

current network to another one, the old user’s IP address is reconfigured to a new
user. A cache log for IP addresses of the departing users in the DNS server [20]
remains active for a specific time and the time lag or delay between the reassigned
IP addresses and the cache log in DNS server creates a significant security issue.
Hence, the attackers may use the cache log of the DNS and modify the DNS with
malware which would violate the originality of the data of new users.

8.2.4.4 Relationship with the third party

The PaaS providers offer web services of different vendors along with their program-
ming language. Cloud provider sometimes subcontracts a third party for data backup
in order to recover data in case of adversity. So the use of multitenant architecture
and flaw in either one may affect the other, and the interconnection complexities
raise many security issues. There must be some mechanism or agreed-­upon policies,
i.e., SLA among the CSPs, subcontracted parties and cloud users to counteract those
issues which can be a security threat for stored data.

8.2.5  The threat management techniques

8.2.5.1 Cryptography
Cryptography is meant as the nonfeasibility of breaking the encryption system and
unable to compute the information about exchanged messages. The main aim of
cryptography is to safeguard the data over an insecure communication channel.
Several techniques are used in cryptography to encrypt and decrypt data. The differ-
ent cryptographic techniques are discussed with their relative advantages and disad-
vantages in the following paragraphs.

8.2.5.1.1  The private key cryptography

Suppose party M needs to send a secret message m to party N which should not be
intercepted by a third party. The Private Key Encryption [21] was the traditional
solution to this problem. The two parties (M and N) meet together and decide to
pass the message by agreeing upon a combination of data encryption and decryption
algorithm ‘E’ and ‘D’ and some additional data ‘S.’ The additional information is
kept secret between M and N. Now, the intruder may know the ‘E’ and ‘D’ algo-
rithm but not aware of the secret key ‘S.’ When M wants to send a communication
‘m’ over an uncertain channel, M encrypts the plaintext message ‘m’ by computing
the cipher text c=E(S, m) and sends c to N. After receiving c, N decrypts the c by
computing m=D(S, c). The intruder does not know S and hence unable to decrypt c
to get the message m. This mechanism is useful for a small amount of data or mes-
sage and if the data are handled by two users only. Now, in the present days, data are
not handled by users alone. A significant amount of data needs to be shared among
millions of different users. Hence a cryptographic key needs to be flexible enough
to handle such big data. A cryptographic algorithm needs to be strong enough to
protect the data. So the private essential encryption technique became obsolete in
modern days.
170  Intelligent network design driven by Big Data analytics

8.2.5.1.2  The public key cryptography

The requirement of M and N to share a secret key ‘S’ in order to encrypt and
decrypt the message ‘m’ was dropped by the development of the Public Key
Cryptography [21]. In this method, the receiver N can publish authenticated
information which is called Public Key. The receiver issues the Public Key to
anyone who wants to send information, including the intruder party. There is an
encryption algorithm containing a Public Key which whoever can recite can trans-
mit an encrypted message to a person whom the sender did not meet personally.
The receiver, by using his secret private key, can decrypt the sender’s message at
his convenience. A trapdoor function is included in the public key encryption algo-
rithm. A trapdoor mathematical function is a single-­side function which contains
some trapdoor or secret data. The receiver alone can retrieve the information to
decrypt the sender’s encrypted message. Diffie and Hellman introduced it in 1976
in the seminal work.
Suppose, M needs to deliver a message m to N,
M needs to compute ƹ (f, m)= f(m)= c and sends to N, where c is the encrypted
message.
To decrypt c=f (m),
N needs to compute fˉˡ (c)= fˉˡ (f(m))= m. f is a trapdoor function.
The public key cryptography is suitable for multiuser systems. Though the pub-
lic key encryption is more secured than the private key, in practical cases, the secret
key needs to be transmitted over the communication channel. Hence, the discovery
of the secret key by the adversary during transmission may cause a threat. One
significant advantage of pubic key encryption is that it can facilitate the digital sig-
natures for authentication, but sharing of secret keys requires trust on the opposition
party that he should not compromise the secrecy of the secret key.

8.2.5.1.3  Data encryption standard with the fixed message

The Data Encryption Standard (DES) [21] is a symmetric block cipher which was
discovered by the National Institute of Standard and Technology (NIST) (Figure 8.7).
Suppose there is a fixed 64-­bit message m that defines the function f(K)= DESĸ (m)
which takes a key K to 64-­bit output f(K). This seems like a one-­way function and
uses a Feistel structure of 16 rounds.
The DES is nowadays considered to be insecure [22] because of a possible brute
force attack in 2008 due to its small key length. The same is superseded by Advance
Encryption Standard (AES) in recent days.

8.2.5.1.4  The Advance Encryption Standard

In AES [21], a substitution-­permutation network is used. Its origin is Rijndael cipher.
The Rijndael cipher uses a different key and block size. It does not use the Fiestel
network, which is used in DES. The AES utilizes a fixed block and set size of 128
bits. A 128, 192 or 256 bits key size is used. The key size is any multiple of 32 bits
with a minimum 128 and a maximum 256 bits. It conducts on a 4 × 4 column-­major
order matrix of bytes which is termed as the state.
 Emerging security challenges in cloud computing: an insight  171

Figure 8.7    DES operation

For instances, suppose there are 16 bytes, starting from x0, x1, x2, x3 … to x15.
These bytes can be represented as the following:
2 3
x0 x4 x8 x12
6 7
6x1 x5 x9 x13 7
6 7
6x x x x14 7
4 2 6 10 5
‍ x3 x7 x 11 x15 ‍

The key size identifies the number of replications of transformation rounds which
translate the input message, i.e., the plaintext, in the final outcome, i.e., the cipher
text. The AES operation can be illustrated as follows (Figure 8.8):
The key length decides the number of rounds that an AES operation should
have. It performs ten sets for 128-­bit keys, 12 sets for 192-­bit keys and 14 sets for
256-­bit keys.
Each set of AES again has four subprocesses (Figure 8.9):

1. SubBytes
2. ShiftRows
172  Intelligent network design driven by Big Data analytics

Figure 8.8    AES operation

3. MixColumn
4. AddRounKey

The biclique attack is one kind of Meet-­In-­The-­Middle (MITM) technique. It

uses a biclique arrangement to expand the number of possible attack sequences by
the MITM attack. In AES, the biclique attack is best known as single key attack. The
attack has 2^126.1 as computational complexity, and hence, it is known a theoretical
attack that is based on security of the AES has not been cracked till now. Care must
be taken for correct implementation of AES and acceptable key management policy.

8.2.5.2 The RSA algorithm

The RSA algorithm [22] is the public key encryptosystem used for encryption of
message and secure data transfer in a network. In this cryptosystem, the key is pub-
lic, i.e., open to all. The decryption key is kept private which is called the private
key. The RSA is drawn from the initial letters of the surname Ron Rivest, Adi
Samir and Leonard Aldeman, the inventors of the algorithm. The RSA is the most
evident algorithm, and it can be applied to safeguard the data in the cloud [23].
The RSA algorithm comprises four primary phases, namely: (1) key generation,
(2) key distribution, (3) encryption and (4) decryption.
The RSA algorithm deals with two keys, namely, the public and a private key.
To encrypt a message, public key is used whereas to decrypt the message private
 Emerging security challenges in cloud computing: an insight  173

Figure 8.9    AES round operation

key is used. Here, we find three big positive integers a, p and d so that the integrated
exponentiation for all m:
(m^a)^p mod d = m

Even knowing a, d or even m, it is very tough to find p.

Modular exponentiation: It is exponentiation calculated over a modulus. It is
useful for public-­key cryptography. It computes the residual when an integer b is
put up to the eth power (the exponent), i.e., be is divided by appositive integer m (the
modulus). The modular exponentiation c is calculated as c ≡ be (mod m).

8.2.5.2.1  Key generation of RSA

The key generation is a crucial part of the RSA algorithm. To generate the key, two
prime numbers are selected distinctly. The prime numbers are of same magnitude
and selected randomly. Now compute n=p.q. For both the private key and public
key n is used as modulus and its length is expressed in bits which are termed as key
length. Calculate φ(n)= φ(p). φ(q)=(p-­1)(q-­1)= n- (P+q-­1). Find out e ≡ d−1 (mod
φ(n)). In simple terms, it can be stated that for a given d.e ≡ 1(mod φ(n)), take out
the solution for e. With the length of e being short and small Hamming weight, more
efficient encryption can be achieved.
174  Intelligent network design driven by Big Data analytics

Here, e and d are released as public key exponent and private key exponent,
respectively. Public key is generated by merging the public exponent e and modu-
lus n, similarly for private key generation, modulus n and private exponent d are
merged.

8.2.5.2.2  The RSA key distribution

If Tom intends to convey a message to Jerry, Jerry needs to transmit public key (d,
a) so that Tom can transmit the message via a reliable but may not a secure channel.
Jerry does not share the private key.

8.2.5.2.3  Encryption
Suppose Tom wants to send message M to Jerry. First, Tom needs to turn the mes-
sage M in integer m, in a manner 0 ≤ m< d and gcd (m, d)=1 by utilizing a padding
scheme which is a prearranged reversible protocol. Tom then calculates the cipher
text c using Jerry’s public key a corresponding to c=m^a mod d. Even for 500-­bit
numbers, this can be done efficiently using modular exponentiation. Tom then sends
c to Jerry.

8.2.5.2.4  Decryption
After receiving the message, Jerry can now recover the message m with the private
key exponent p by calculating c^p = (m^a) ^P=m mod d. Jerry finds the original
message M using m and reversing the padding scheme.

8.3 Review on existing proposed models

8.3.1  SeDaSC
Ali et al. [14] proposed a model on Secure Data Sharing on Cloud (SeDaSC) titled
‘SeDaSC: Secure Data Sharing in Cloud.’ In their model, the user file is encrypted
by a single encryption key called the master key. After encryption of the file, the
master key is then divided into two different key shares for every user. One key part
is possessed by the user, which keeps away the intruder (insider threat) from the user
data and another key part is retained from a reliable third-­party body which is termed
as Cryptographic Server (CS). The master key is then deleted permanently. The
key part alone cannot decrypt user data. It needs to generate the master key again
with the help of the two key parts to decrypt the user’s file. See Figure 8.10.

8.3.1.1 Working of the ‘SeDaSC’ model

There are three main entities in ‘SeDaSC’ model, namely: (a) users, (b) CS and (c)
the cloud. In the primary phase, a user or data holder loads the data, list of users who
are going to use the data and the parameters which are needed to generate an Access
Control Lists (ACL) in the CS. The CS then being a reliable third party takes all the
responsibilities of access control, key management, encryption and decryption of
the data/file. The CS generates a master key and encrypts the data with the produced
 Emerging security challenges in cloud computing: an insight  175

Figure 8.10    The basic idea of SeDaSC methodology

master key. To prevent the regeneration of the original master key, the master key is
divided into two parts. Secure overwriting is used to delete the original master ley.
One part is the CS retains the key along with the ACL list maintained by the CS and
the other part is being sent to the users in the circle. The encrypted file is stored in the
cloud by the CS on behalf of the users. A CSP maintains the cloud storage [24–32].
The user, when wants to retrieve the data, sends an access request to the CS. The
CS, after receiving the request, downloads the required user’s file and asks for the
key part available with the user. After receiving the user key part, it authenticates
and reconstructs the master key with the help of the user’s key, and CS maintained
portion of the key for that particular user. The user’s file is then decrypted with the
master key newly generated and sent back to the user who requested the data. If a
new member joins with the existing group, the new user ID is added to the ACL, and
two parts of the key are again generated. For the member who is leaving the group,
his identification is deleted from the existing ACL. There is no possibility of access-
ing the data by the departing member, as he is having only the user part of the key
[33–40]. The ‘SedaSc’ model also suggests that the frequent encryption/decryption
is not required in the event of any change in the group membership.
The ‘SeDaSC’ model claims that the methodology can be utilized in mobile
cloud computing in supplement to existing conventional cloud computing since the
cryptographic server performs compute-­intensive operation.

8.3.1.2 Discussions on SeDaSC model

Though the SeDaSC model seems to be muscular in terms of safety and privacy of
data, it lacks in addressing so many issues. The model itself is a complex one. The
computational burdens in this model make it unhealthy for handling millions of data
176  Intelligent network design driven by Big Data analytics

and keeping user’s data separate from other users in the network. The following
issues need to have greater attention to upgrade the security issues for the data.

1. The cryptographic server (CS) :

1. To maintain a CS in the cloud environment is a considerable task
and two-­step generation of user key will impose an intensive
computation cost and an overburden increase in pay-­per-­use defi-
nitely would discourage the cloud users from storing their data.
2. The administrators of the CS may not be trustworthy. The tainted
administrators may retain the user’s portion of the key, and the
data security and privacy may be violated.
3. A duplicate ACL list for the outgoing members of the group may
exist in the server as a mirror image and can be used by the de-
parting members as they are already holding the user portion key.
4. The CS may not receive a response due to unreachability/with-
drawal/denial of service at any point in time, and then whole data
will be lost. There is no mechanism to recover the data as the user
alone cannot recover the data without the CS portion of the key.
Moreover, the encryption and decryption of data are done by the
CS only. So the users have no control over their data.
2. The sharing of keys within a multitenant cloud environment is not recommended.
3. The CSP needs to maintain user’s records along with the CS records and an
additional network link with the CS, which makes the cloud environment a
critical mesh.

8.3.2  The ‘SecCloud’ protocol

Lifei, Haojin Jhu et al. [41] introduced a privacy cheating discouragement and
secure computation auditing protocol which is named as ‘SecCloud.’ They claimed
that the proposed ‘SecCloud’ is the first protocol that bridges between protected
storage and protected computation auditing into the cloud. It achieves the privacy
cheating discouragement through specified verifier signature, batch verification and
probabilistic sampling techniques. They also tried to develop a ‘practical secure-­
aware cloud computing experimental environment’ which is named ‘SecHDFS,’ a
test platform to employ ‘SecCloud.’

8.3.2.1 The working of the ‘SecCloud’

The model considers a general cloud computing model which is having several
cloud servers S1 to SN. The servers are in control of one or multiple CSPs. A mobile
phone or a laptop (called as cloud users or CU) which is having lesser storage is con-
nected to CSP to avail the resources (computation and storage) of the cloud. It is also
assumed that there are verification agencies or VAs trusted by CUs and responsible
for auditing the storages and computing of the CSPs. VAs are assumed to be having
more powerful computing capability than CUs. See Figure 8.11.
 Emerging security challenges in cloud computing: an insight  177

Now, if any adversary ‍A ‍could corrupt a small group of cloud servers and affect
these servers to introduce numerous cheating attacks as below:

a. Storage cheating attacks: when the adversary modifies the store data to negoti-
ate data integrity or discloses the confidential data to purchase interest or both.
b. Computing cheating attack model: The attack can be for data computation secu-
rity of the cloud. The attackers can leave the cloud with erroneous computing
for the users, yet the data seem to be original.
c. Privacy cheating attack model: The attackers can reveal or leak the data to the
public for business competition. If the data are not encrypted, i.e., plaintext data,
the data can be sold out by the adversary.

To protect the valuable data from the attackers, the ‘SecCloud’ provides two
concepts as follow:

a. Secure computation confidence (SCC): to specify the trust level of computation

security
b. Secure storage confidence (SSC): to improve the confidence level of storage
security

The SCC shall be defined as ǀF’ǀ/ǀFǀ and SSC is defined as ǀX’ǀ/ǀXǀ,

where F is the total critical computation request, F' is the computation carried
out precisely, X is the requested data set and X’ is the honestly returned data set.
If the values of SCC and SSC are both equal to 1, then the cloud computation and
cloud storage are regarded as trusted. Otherwise, they are semitrusted.
Based on the above probability, the report suggests two definitions which would
safeguard the cloud storage and calculation in the same way.
The protocol is able to achieve the following goals:

Figure 8.11    Cloud computing architecture in SecCloud protocol

178  Intelligent network design driven by Big Data analytics

a. Data storage security by the effective auditing of CU and VA.

b. Data computation security by VAs verification and auditing by CU and VA [42,
43].
c. Privacy cheating discouragement because of verification by a designated party
and discouraging CSPs for revealing privacy of CUs data even if the attackers
attack the servers.

8.3.2.2 Discussion on ‘SecCloud’ protocol

Before running the proposed protocol, there is a need for system set-­up step by the
System Initialization Operator (SIO) which generates the system parameters params
and the master secret key ‘s.’ There is another offline user registration step in which
user needs to register to SIO, and after submitting the customer ID to SIO, the user
receives system parameter params and a secret key.
The proposed protocol performs the following four steps:

a. The proposed protocol requests for a storage space to the CSPs and the CSP
allocates a space by returning a space index i for the message to be stored.
b. The CUs need to sign each transmitted message block to enable the VAs for
auditing.
c. CUs carry out data encapsulation for precomputing a session key by using the
Bilinear Diffie-­Hellman method. The CUs then send the data encrypted by the
session key and corresponding signature pairs to the cloud for storage.

When the data are needed to be received, then the CSP decrypts the packet by
using its session key to recover data signature pairs and check the signature for data
authenticity by VAs using its secret key. The authority for checking the signatures
is held by the CSP and VA only, and hence, it is claimed that the data are secured
and protected.
Though the ‘SecCloud’ protocol claims to be robust and reliable, its shortcom-
ings can be found as below:

a. The computing and transmission overhead is the primary concern here. The
computing and transmission of encrypted data are very much involved and take
much time which may cause time-­out for access to the servers.
b. Here in this proposed model, we are again going to be dependent on a third
party, i.e., Verification Auditors (which may reveal the valuable user’s data and
cause a threat. Verification auditors need to be persuaded that the cloud servers
use the data on the exact location so that the cloud server’s cheating behavior
should not be detected.
c. In some cases, the users may face a denial of service due to the fact that verifica-
tion auditors may become unresponsive due to connection loss with the users
or servers.
 Emerging security challenges in cloud computing: an insight  179

Figure 8.12    The cloud information accountability framework

8.3.3  D
 ata accountability and auditing for secure cloud data
storage
Prassanna et al. [44] and the team have evaluated the structure for data account-
ability and reviewed cloud user data in the distributed cloud. They have proposed
a mechanism that is technically monitoring any data access collected in the cloud
with reviewing mechanism. The mechanism also suggests privacy – maintaining
public auditing model for customer data that uses the cloud storage services in cloud
computing.

8.3.3.1 Working of the ‘data accountability and auditing’ model

In cloud information processing (Figure 8.12), the model uses the principle of object-­
oriented technique. There is a component logger that is connected to the cloud cli-
ents’ data. The extraction is performed when the user data are retrieved from the
remote server, and logging information is replicated with accessed data. The user’s
current processing of data and the component are accounted and kept a replica in the
component log harmonizer. The second logger establishes the essential centralized
element to enable the user to log in it.
The data user can come to know that whether the data are accessed by some-
body else by the log harmonizer which keeps track of logging credentials of every
data accessories.
The model again proposed a third-­party auditing authority who is responsible
for auditing the data accessibility. Vulnerability exists due to internal threats.
180  Intelligent network design driven by Big Data analytics

Here, it is noticed that most of the models are relying on a third party which
is the leading cause of concern in the security issues of cloud computing. In this
advanced era of technology, it is not advisable to rely on a third party.

8.4 Conclusion and future prospectives

The storage, virtualization as well as the various virtualization techniques and the
network in use encompass a more significant security concern in a cloud computing
environment. The security of user’s data needs to be ensured at the highest level.
Due to the complex architecture of the cloud, it is a big challenge to ensure the
security of data at all levels in the cloud environment. Interconnections are complex
among different architectures of CSPs. PaaS providers hire SaaS vendors for their
service and also sometimes a third party for the backup of the user’s data. The dif-
ferent providers must agree upon an SLA to secure the user’s data. Users must be
informed that what levels of standard securities are available with the CSPs and how
their data are being stored.
This chapter discusses various security issues and challenges that are present
in different cloud models and cloud environment. It gives a complete idea of other
threat management techniques available to encounter security issues and challenges.
Also, the chapter reviews the different proposed approaches published by various
journals. The report explains an entirely different process (secure key transmission
model) to enhance the security of the above issues and challenges, discussions and
comparisons with the existing models.

References

[1] Cloud computing [online]. 2020Wikipedia. Available from https://

en.wikipedia.org/w/index.php?title=Cloud_computing&oldid=987774983
[Accessed 9 Nov 2020].
[2] Buyya R., Pandey S., Vecchiola C. Cloudbus Toolkit for Market-­Oriented
Cloud Computing. Cloud Computing; Berlin, Heidelberg; 2009. pp. 24–44.
[3] IBM Software. 2020IBM [online]. Available from https://www.ibm.com/​
products/software​h ttps://​w ww.​i bm.​c om/​p roducts?​t ypes[0]=software
[Accessed 9 Nov 2020].
[4] Ren Y., Tang W. ‘A service integrity assurance framework for cloud com-
puting based on MapReduce’. IEEE 2nd International Conference on Cloud
Computing and Intelligence Systems; Hangzhou, China, 30 Oct-­1 Nov; 2012.
pp. 240–4.
[5] Karun A., Karat C. ‘A review on hadoop—HDFS infrastructure exten-
sions’. IEEE Conference on Information & Communication Technologies;
Thuckalay, India, 11-­12 Apr; 2013.
[6] Mell P., Grance T. The NIST definition of cloud computing. 800-­145. National
Institute of Standards and Technology, Gaithersburg, MD: NIST SP; 2011.
 Emerging security challenges in cloud computing: an insight  181

[7] Fernandes D.A.B., Soares L.F.B., Gomes J.V., Freire M.M., Inácio P.R.M.
‘Security issues in cloud environments: a survey’. International Journal of
Information Security. 2014, vol. 13(2), pp. 113–70.
[8] Security guidance for critical areas of focus in cloud computing v4.0 [on-
line]. cloud security alliance. Available from https://cloudsecurityalliance.
org/artifacts/security-guidance-v4/
[9] Understanding-­the-­Cloud-­Computing-­Stack [online]. Available from htt-
ps://www.diversity.net.nz/wp-content/uploads/2011/03/Understanding-the-​
Cloud-Computing-Stack.pdf [Accessed 9 Nov 2020].
[10] Cloud Security and Privacy [online]. Available from https://www.oreilly.com/​
library/view/cloud-security-and/9780596806453/ [Accessed 9 Nov 2020].
[11] Keene C. What Is Platform as a Service (PaaS) [online]?Available from
http://www.keeneview.com/2009/03/what-is-platform-as-service-paas.html
[Accessed 9 Nov 2020].
[12] Hashizume K., Rosado D.G., Fernández-­Medina E., Fernandez E.B. ‘An
analysis of security issues for cloud computing’. Journal of Internet Services
and Applications. 2013, vol. 4(1), p. 5.
[13] Jansen W.A. ‘Cloud hooks: security and privacy issues in cloud computing’.
44th Hawaii International Conference on System Sciences, IEEE; Kauai, HI,
2011. pp. 1–10.
[14] Ali M., Dhamotharan R., Khan E., et al. ‘SeDaSC: secure data sharing in
clouds’. IEEE Systems Journal. 2017, vol. 11(2), pp. 395–404.
[15] Zhang F., Chen H. ‘Security-­preserving live migration of virtual machines in
the cloud’. Journal of Network and Systems Management. 2013, vol. 21(4),
pp. 562–87.
[16] Subashini S., Kavitha V. ‘A survey on security issues in service delivery mod-
els of cloud computing’. Journal of Network and Computer Applications.
2011, vol. 34(1), pp. 1–11.
[17] treacherous-­12-­top-­threats [online]. Available from https://downloads.clou​
dsecurityalliance.org/assets/research/top-threats/treacherous-12-top-threats.​
pdf [Accessed 9 Nov 2020].
[18] Wei J., Zhang X., Ammons G., Bala V., Ning P. ‘Managing security of vir-
tual machine images in a cloud environment’. CCSW ’09: Proceedings of
the 2009 ACM Workshop on Cloud Computing Security, ACM; Chicago, IL,
2009. pp. 91–96.
[19] Ohlman B., Eriksson A., Rembarz R. ‘What networking of informa-
tion can do for cloud computing’. 18th IEEE International Workshops on
Enabling Technologies: Infrastructures for Collaborative Enterprises, IEEE;
Groningen, Netherlands, 2009. pp. 78–83.
[20] Zhang L., Zhou Q. ‘CCOA: cloud computing open architecture’. 2009 IEEE
International Conference on Web Services, IEEE; Los Angeles, CA, 2009. pp.
607–16.
[21] Goldwasser S., Bellare M. Lecture notes on cryptography. introduction to
modern cryptography [online]. 2008. Available from https://cseweb.ucsd.​
edu/~mihir/papers/gb.pdf
182  Intelligent network design driven by Big Data analytics

[22] Rivest R.L., Shamir A., Adleman L. A method for obtaining digital signatures
and public-­key cryptosystems, communications of the ACM. Vol. 21; 1978.
pp. 120–26.
[23] Wei J., Zhang X., Ammons G., Bala V., Ning P. 'Managing security of virtual
machine images in a cloud environment, ACM 91-­96'.2009, vol. 6(1).
[24] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wire-
less sensor network’. Fifth International Conference on Communication
Systems and Network Technologies (CSNT), IEEE; Gwalior, India, 2015. pp.
194–200.
[25] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distributed lo-
calized wireless sensor networks’. International Conference on Issues and
Challenges in Intelligent Computing Techniques (ICICT); Ghaziabad, India,
7-­8 Feb; 2014.
[26] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON), IEEE; Mathura, India, 2021. pp. 1–6.
[27] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems; Singapore: Springer; 2020.
[28] Kumar S., Trivedi M.C., Ranjan P., Punhani A. Evolution of Software-­Defined
Networking Foundations for IoT and 5G Mobile Networks. Hershey, PA: IGI
Publisher; 2020.
[29] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[30] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[31] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
Personal Communications. 2021, vol. 10(3).
[32] Kumar S., Cengiz K., Trivedi C.M., et al. ‘DEMO enterprise ontology with
a stochastic approach based on partially observable Markov model for data
aggregation and communication in intelligent sensor networks’. Wireless
Personal Communication. 2022.
[33] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering. 2019, vol. 8(6), pp. 2278–3075.
[34] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
 Emerging security challenges in cloud computing: an insight  183

classifier’. International Journal of Computational Vision and Robotics.

2020, vol. 10(6), pp. 505–21.
[35] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[36] Singh P., Bansal A., Kamal A.E., Kumar S.. (eds.) ‘Road surface qual-
ity monitoring using machine learning algorithm’ in Reddy A.N.R., Marla
D., Favorskaya M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and
Energy Sustainability. Smart Innovation, Systems and Technologies. 265.
Singapore: Springer; 2022.
[37] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium, Publisher: The Electromagnetics
Academy; Prague, Czech Republic, 2015. pp. 2363–67.
[38] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
International Conference on Computational Intelligence and Communication
Networks (CICN), IEEE; Jabalpur, India, 2016. pp. 79–84.
[39] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 4th International Conference on Information Systems and
Computer Networks (ISCON), IEEE; Mathura, India, 2019. pp. 250–55.
[40] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. 10th International
Conference on Cloud Computing, Data Science & Engineering (Confluence),
IEEE; Noida, India, IEEE, 2020. pp. 63–76.
[41] Wei L., Zhu H., Cao Z., et al. ‘Security and privacy for storage and com-
putation in cloud computing’. Information Sciences. 2014, vol. 258(4), pp.
371–86.
[42] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementa-
tion of fault tolerance technique for Internet of Things (IoT)’. 2020 12th
International Conference on Computational Intelligence and Communication
Networks (CICN); Bhimtal, India, 25-­26 Sep; 2020. pp. 154–9.
[43] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and iot for
smart road traffic management system’. IEEE India Council International
Subsections Conference (INDISCON), IEEE; Visakhapatnam, India, 2020.
pp. 289–96.
[44] Prassanna J., Punitha K., Neelanarayanan V. 'Towards an analysis of data ac-
countability and auditing for secure cloud data storage'. Procedia Computer
Science, Elsevier. 2015, vol. 50(4), pp. 543–50.
This page intentionally left blank
 Chapter 9
Factors responsible and phases of speaker
recognition system
Hunny 1 and Ayush Goyal 2

The method of identifying a speaker based on his or her speech is known as automatic
speaker recognition. Speaker/voice recognition is a biometric sensory device that
recognizes people by their voices. Most speaker recognition systems nowadays are
focused on spectral information, which means they use spectral information derived
from speech signal segments of 10–30 ms in length. However, if the received speech
signal contains some noise, the cepstral-­based system’s output suffers. The primary
goal of the study is to see the various factors responsible for improved performance
of the speaker recognition systems by modeling prosodic features, and phases of
speaker recognition system. Furthermore, in the presence of background noise, the
analysis focused on a text-­independent speaker recognition system.

9.1 Study of related research

Many researchers have labored to develop various methods and algorithms for ana-
lyzing speech signals and simplifying speech and speaker recognition processes. In
Reference [1], recent audiovisual (AV) fusion research is summarized. In References
[2–6], a method is proposed to solve the issue of visual speech recognition.
The fusion of multiple biometric characteristics for identity authentication has
shown strong benefits as compared to conventional systems based on unimodal
biometric attributes. By combining visual speech and face information simultane-
ously, a new multimodal verification approach is explored in this study. Unlike face
authentication, the proposed method uses visual speech lip movement features,
which can reduce the risk of being duped by a fake face picture. To extract features
of the face and visual expression, a Linearity Preserving Projection transform and a
Projection Local Spatiotemporal Descriptor are used.

KIET Group of Institutions, Delhi-­NCR, Ghaziabad, India

1

Department of Electrical Engineering and Computer Science, Texas A&M University – Kingsville,
2

Kingsville, TX, USA

186  Intelligent network design driven by Big Data analytics

To combine the multisource biometric features, an Extreme Learning Machine-­

based fusion method is used in the corresponding score level to generate a fused
score for the final verification. Experiments on the OuluVS database have shown
that the proposed approach is capable of producing very good results.
Auxiliary loss multimodal GRU model [7] is divided into three sections:

• Feature extraction
• Data augmentation
• Fusion and recognition

Direction of arrival (DoA)-[8], TDoA or interaural time difference (ITD)-[9–

12], ILD-[13–16], and HRTF-­based processes are the most popular passive sound
source localization methods.
A number of localization methods have been proposed for multiple signal esti-
mation, covariance approximation, and signal parameter estimation using rotational
invariance techniques [17]. Methods based on ILD, on the other hand, are often
valid for a single dominant sound source [high signal-­to-­noise ratio (SNR)] in case
of single dominant sound source [13–16], to name a few.
The minimum number of microphones available for ILD- or TDE-­based meth-
ods is three [9–12] and [18–26]. Only HRTF-­based techniques can be used to calcu-
late the arrival angle. GCC with an ML estimator [20] is a widely used tool for TDE.
Later References [18, 19, 21, 25, 27] suggested a slew of methods for developing
GCC in the presence of noise.
For high accuracy robust speaker localization, a new PHAT-­based approach
known as steered response pattern-­phase or power-­phase transform was recently
proposed [26]. The disadvantage of this method is that it takes longer to process than
PHAT [22], as it necessitates the identification of a large number of user locations.
References [13, 28, 29] used ILD to pinpoint the location of sound sources. While
these studies used ILD-­based methods to locate a source, Reference [30] used a
pair of microphones to try 2D sound source localization using TDE- and ILD-­based
methods simultaneously. Reference [15] addresses a typical scenario of 3D localiza-
tion with four or more sensors to boost source position accuracy. Sen and Nehorai
recently used the Brown–Duda HRTF model to model multipath delays close to the
sensor [23], and frequency-­dependent head-­shadow effects, in order to investigate
3D direction finding using auditory system of human with only two sensors [31].
In the last decade, several papers have attempted to use HRTF in combina-
tion with TDE to estimate elevation and azimuth angle of arrival [32]. Reference
[33] concentrated on a straightforward beamforming technique. The Taylor series
expansion [34] linearization solution has a high computational complexity, iterative
processing, and needs a tolerable initial estimation of the location for convergence.
Reference [9] suggested a method based on the assumption of a distant source. In the
small error field, Reference [35] proposed a solution that can achieve the Cramer–
Rao lower bound. When sensors are spread at random, the situation becomes more
complicated. The intersection of a series of hyperbolic curves found by TDoA
 Factors responsible and phases of speaker recognition system  187

estimates using non-­Euclidean geometry [36]. Reference [37] determines the emit-
ter location in this case.

9.2 Phases of speaker recognition system

In the below-­mentioned subsections, previous research work on recognition system

of speaker is analyzed and addressed for each developmental phase.

9.2.1  Speaker database collection

This is where the development of a speaker recognition system starts. The develop-
ment and evaluation of speaker recognition systems rely heavily on the basic speech
database [38]. Changes in the field of speaker recognition could be influenced by the
availability of speech databases.
When collecting speaker database for biometric speaker recognition, the most
important factors to consider are session and channel variability, but other variables
are also important to consider. Figure 9.1 depicts some of the most critical variables.

Figure 9.1    Various factors for a robust speaker database collection

188  Intelligent network design driven by Big Data analytics

9.2.1.1 Variability of session
Session variability applies to a phenomenon that causes differences between two
recordings of the same speaker [39]. To put it another way, two different samples of
speech recorded by the same person are unidentifiable by the computer [40].
Intersession variability is caused by a number of factors as follows.

9.2.1.1.1  Channel of transmission

It is a big part of why there is so much variation between sessions. This occurs due to dif-
ferences in the transmission channels' characteristics. Despite the fact that speech signals
obtained from voice transmitted over mobile phones (i.e., wireless channel) may differ
from speech samples transmitted over landlines (i.e., wired channel), speech signals col-
lected from voice transmitted over landlines (i.e., wired channel) may differ.

9.2.1.1.2  Characteristics of transducer

“Transducer” means a system that converts one signal into another. In the case of
speech, a microphone is a transducer that transforms an audio signal into an electric
signal. After that, the electrical signal is transformed into a speech sample, which is
then saved. The type of microphone used has an effect on the speech sample’s accu-
racy. The speaker recognition system cannot correctly identify the target speaker
when different microphones are used in training and testing equipment [41].

9.2.1.1.3  Noise of environment

The amount of noise in different recording sessions can also vary, which may cause
intersession variability. Due to number of things such as traffic, people talking near
the recording site, and nearby industry, noise may be triggered [42].

9.2.1.1.4  Variability of intraspeaker

Same speaker’s speech samples may have intersession variability because the acous-
tic characteristics of the speech sample are primarily dependent on the speaker’s age,
and psychological health, emotional state, and other factors. As a result, if there is a
significant time delay between training samples and testing recordings, the features of
the same speaker’s voice can differ significantly. Intersession variability [43] must be
compensated for reliable speaker recognition. Compensation methods such as feature
mapping, feature warping, and others have been introduced [44, 45].
In contrast to biometric speaker recognition, there are a number of other aspects
that must be discussed in forensic speaker recognition. The main causes of these
problems are uncontrolled recording conditions and a lack of background speech
data for comparison. To create a database for a forensic application, the devel-
oper must adhere to certain legal requirements. The law officials imposed some
restrictions for the use of this content, one of which is that the final database cannot
reveal a speaker’s identity [15]. Real phone conversations from real police cases
are included in the database. AHUMADA 3, a speech database recorded in Spain,
is similar. Database Availability and Database Mismatch in the Sense of Forensics
[46] are two major issues in FSR.
 Factors responsible and phases of speaker recognition system  189

9.2.1.2 Gender
Biometric systems use sources like iris, palm-­print, face, hand geometry, and voice,
etc., to recognize individuals [47]. Along with above-­mentioned characteristics,
some ancillary knowledge about the user can be used to build a secure and user-­
friendly biometric device. Height, gender, age, and eye color of the individual person
are examples of ancillary details. These are referred to as “Soft Biometric” charac-
teristics. These characteristics may be continuous or discrete in nature. Gender is a
discrete soft biometric characteristic. Despite the lack of distinctiveness in the use of
soft biometric traits, gender can be used to filter [24] a broad biometric database [48].
Male and female speaker’s voices vary greatly in the shape of some aspects of the
speech signal, allowing them to separate out a huge amount of unnecessary data and
a considerable amount of time is saved.

9.2.1.3 Environment
Researchers have recently become concerned about mismatch among training and
testing environments. Speaker model synthesis [49], factor analysis [50], and feature
mapping [51] are some of the techniques developed to address this problem. Parallel
condition data are needed for accurate speaker recognition because features of the
speech signal are influenced by surrounding conditions. Speech sample collected in
the soundproof setting will have much better quality than the one recorded in a noisy
classroom or library. A database with speech samples collected in various environ-
ments, such as a soundproof space, a noisy classroom, a library, an auditorium, and
a market, should be created for this purpose.

9.2.1.4 Language and region

Speakers in a text-­independent environment are not limited in what they may say for
verification. This is an implicit verification while speaker is engaged in some activ-
ity, i.e., filing a complaint or conversing with a service representative. The majority
of the speaker recognition scheme is bound to a single language [31]. India is a
country with a large number of regional languages, making it a highly multilingual
country. In addition, each region has its own influence on the language and its acous-
tic characteristics. In addition, different dialects of the same language are used in
various parts of India. Pitch, energy, phenomes, and other acoustic characteristics of
English spoken in North India differ from English spoken in South India. As a con-
sequence, it is important to see how multiple languages, as well as a single language
spoken in various areas, affect the system of speaker recognition [52, 53].

9.2.1.5 Instrument of recording
Speech samples may be recorded using a variety of instruments. The quality of the
speech samples is influenced by the methods of recording. A digital voice recorder,
laptop, cell phone, microphone, and long-­distance phone call are some of the devices
that can be used to build a quality database [54].
190  Intelligent network design driven by Big Data analytics

9.2.1.6 Age variability
The issue of aging and variations in speech quality is intertwined. Over time, the
aging effect becomes more pronounced, and the standard of speech becomes more
likely to deteriorate. Because of the aging effect, accuracy of any biometric system
degrades over time [55, 56]. However, the impact of age variability on speaker rec-
ognition has received the attention of sporadic research. The problem of aging in a
speaker recognition system can be solved by database updating at regular intervals.
A better, but more difficult, approach is to automatically adapt aging-­related
changes. The lack of a database is the most significant challenge in developing
such a method. The public access to the longitudinal speaker database, which
spans more than three years, is inaccessible. The key source of variability in the
TCDSA (Trinity College Dublin Speaker Aging) database was ageing, but vari-
ance in speech quality was inevitable for such a long period of time. As a result,
for a long-­term and large-­scale method, the database must contain data from dif-
ferent speakers at different times, or the database should be constantly updated in
this manner.

9.2.1.7 Spoofing
Types of spoofing attacks are: (a) impersonation, (b) speech synthesis, (c) voice con-
version, and (d) replay. Impersonation is when someone attempts to imitate another
person who is a real speaker. The synthetization of the voice for the authentic
speaker is done by using a speech synthesizer to spoof the verification process [57].
Conversion of voice is a method of spoofing the computer in which attacker’s voice
is automatically converted into the voice of a legitimate speaker using a conversion
tool. The goal genuine speaker’s prerecorded speech samples are replayed using a
playing computer, which can be a cell phone, music player, or any other player. As
a result, steps should be taken right from the start, during the compilation of speech
databases, to prevent attacks of spoofing.

9.2.1.8 Whispering
The effect of deliberate speech behavior alteration on speaker recognition is studied,
and a flaw in speaker recognition systems is presented. Whispered speech, a form
of disguised speech created psychologically and/or physiologically, has recently
piqued the interest of researchers [58]. Along with research into its acoustic char-
acteristics, such as formant frequencies, corresponding bandwidth, and endpoint
detection, researchers are also interested in its applications, such as reconstruction,
speaker recognition, and so on. Whispered speech characteristics are:

i. Since vocal cords do not vibrate, exhalation is the source of

excitation.
ii. SNR of whispered speech is low.
iii. When whispering, the enunciator’s psychology is vulnerable [59].
 Factors responsible and phases of speaker recognition system  191

9.2.1.9 Twins
Since 1990, the birth rate of twins has increased by an average of 3% each year,
according to statistics [60]. Even though identical twins account for just 0.2 per-
cent of the world’s population, their numbers are comparable to the populations of
countries such as Greece and Portugal. As a result, a biometric system capable of
accurately distinguishing between identical twins that share the same genetic code is
urgently needed [61]. The results showed that the voice and expression parameters
of the twin pairs differ to varying degrees of similarity and dissimilarity.

9.2.1.10 Different speaking styles and situational mismatch

Unknown speech samples are captured from various sources, most of which are
unknown to the speaker who utters them. As a result, the style of speaking can be
similar to or completely different from that of the stored speech database [62].
A regular telephone conversation, something read by the speaker, or the speech
in which the speaker is shouting over someone is all examples of the style. It is likely
that the recorded speech contains the samples of two and more people (as in a meet-
ing) or that someone is deliberately concealing to cause doubt. This may also be part
of a police-­suspect interview or a telephone information exchange [63].

9.2.1.11 Different stress levels and mental state of the speaker

Speech samples can result in an error if the speaker is in a different mental state at
the time of two samples to be compared. For example, if one speech sample was
taken when the speaker was in a normal mental state and another speech sample
was taken when the speaker was ill mentally or in the influence of a sedative drug.
The intraspeaker variability in both samples can be very high. Different levels of
stress may also trigger this, regardless of whether the stress is physical, mental, or
emotional [63].

9.2.1.12 Sparse background data

Due to legal issues with forensic database selection, the history database needed
for the creation of speaker recognition model is scarce [64]. As a result, researchers
have access to a limited amount of forensic data. No such data are available in the
Indian context to our knowledge. A number of researchers have tried their hardest to
compile databases with various samples approximate close to the practical data as
possible, but not able to solve all of the above-­mentioned issues.
Table 9.1 contains a list of few of the most well-­known databases from all over
the world. In India, some databases are also being created. The Chandigarh-­based
Center for Forensic Science Laboratory (CFSL) has created one database. To solve
the issue of dialect variance and disguise as well as channel heterogeneity, they
gathered a collection of ten different languages. One more database was created for
a robust speaker recognition system that is not language-­based. Captured samples
in Hindi, Marathi, and Urdu to provide a diverse data for cross-­lingual and mul-
tilingual speaker recognition as well as captured 180 Hindi, 200 Marathi, and 70
192  Intelligent network design driven by Big Data analytics

Table 9.1    Some available database for speaker recognition

No. of
speakers
S. Name of Language Duration/ Type of data/
no. database used size M F Text Used EER References

1 AHUMADA I, Spanish – 150 250 Sentences/ 0.5 [65]

II, III Digits/
Words
2 ESTER- I, II French 100 Hours – – News, Debates, 1.1 [66]
Tv Shows
3 ETAPE French 30 Hours – – News, Debates, – [67]
Tv Shows
4 AUS-­TALK English 3,000 Hours 1,000 Story/ – [68]
Sentences/
Digits/
Words
5 REPERE French 60 Hours – – News, Debates, – [69]
Tv Shows
6 WHISPE Spanish 5,000 words 5 5 Words – [70]
7 CIVIL-­CORPUS Spanish 20 Hours 28 32 Sentences/ – [71]
Digits/
Words
8 NFI-­FRITS Dutch 4,188 604 Conversations 12.1 [72]
Conversation
9 FABIOLE French 310 utterances 130 News, Debates, 2.5 [73]
Tv Shows

Urdu speakers who record these samples by using various equipment for channel
variability.
They also gathered databases for identical twins for the requirements of
such information by forensic scientists. Another effort, using five separate net-
works in parallel, was made to build a database of 200 speakers in English
and many Indian languages in the soundproof as well as noisy places. These
databases are commonly used for biometric and forensic applications, and they
still have some flaws. Many research issues, such as databases for crying and
shouting voices, distance from the microphone, age variability, and spoofing,
remain unsolved despite these efforts.

9.2.2  Feature extraction

The second step is to extract features from various speech signals. Speech has many
characteristics, but a few characteristics are necessary to distinguishing various
speakers.
Following are the optimal characteristics of the ideal function:

• It should be easily extracted from the speech signal.

• It should not be influenced by age variations.
• It should be used consistently and naturally in voice.
 Factors responsible and phases of speaker recognition system  193

• It must be impossible to imitate or mimic.

• It should be robust for distortion and noise.

Multiple features must be used for speaker recognition, as all the above-­
mentioned characteristics are not fulfilled by a single feature. At the same time, the
number of features considered for processing and recognition should be limited, as
few techniques can handle high-­dimensional data [11].
These characteristics have disadvantages as well, such as being less discrimina-
tive and easily mimicked. High-­level functionality often necessitates a more com-
plex framework. As a result, no function can be considered the best for recognition,
and feature selection is a trade-­off between robustness, discriminative property, and
device implementation viability.

9.2.2.1 Short-term spectral features

Because of articulatory gestures, sound signals are extremely nonstationary. As
a result, speech signal is broken into short frames for a pseudo-­stationary signal.
These short frames have a length of about 20–30 ms, and for extraction of spec-
tral features, these frames can be considered stationary during this period. Glottal
voice source causes a downward sloping spectrum, resulting in very low intensity
at higher frequency range. Therefore, in order to raise the higher frequencies, these
frequencies must be pre-­emphasized [74].
The magnitude continuum is thought to be more significant and is usually
retained for further processing on the basis of assumption that the phase spectrum
contains only minor perceptual information. The “spectral envelop of the contin-
uum” contains a vast information in speaker identification. The spectral envelop is a
global form of DFT’s magnitude spectrum. Lower frequencies receive more band-­
pass filters in order to reflect them with greater resolution [75].
Transformations are now used to reduce the dimensionality of sub-­band energy
values, which was previously used as features directly [76].
Multiple features used in combination, on the other hand, will lower the recog-
nition error rate. In reality, any function from MFCC, LPCC, LSF, and PLP can be
used depending on the application.

9.2.2.2 Voice source features

These characteristics bear speaker-­specific data such as glottal pulse shape, fun-
damental frequency, and so on. The efficiency of the speech is determined by the
parameters [77–90]. Close-­phase covariance analysis can be done if the vocal folds
are closed. Other methods have employed parameters such as residual phase, glot-
tal flow model, cepstral coefficient, higher order statistics, and others. Voice source
features are less reliant on phonetic content, while features of vocal tract are far
more reliant on phonetic factors and, as a result, require extensive phonetic cover-
age, which necessitates a significant amount of data for training and testing. It is also
noticeable that combining these two features will increase accuracy.
194  Intelligent network design driven by Big Data analytics

Figure 9.2    Temporal discrete cosine transform

9.2.2.3 Spectro-temporal features
Two spectro temporal details, i.e., formant transitions and energy modulation, may
be used to extract a lot of speaker-­specific detail.
 2  To provide temporal details to
the functions, the delta ‍   and
‍ double delta ‍   coefficients,
‍ which are first- and
second-­order derivative estimates, can be used. The successive feature vector coef-
ficient’s time difference is used to calculate these coefficients and then combined
with the original coefficients. If thenumber
 of original coefficients is n, then the total
number of coefficients with ‍‍ and ‍ 2 ‍coefficients will be 3 n. For each frame, the
process is repeated.
Figure 9.2 depicts the temporal discrete cosine transform process. Performance
can be improved by integrating cepstral and temporal features instead of cepstral
system alone, but the change was minor, and further study is needed before it can be
used. Speaker recognition may also use modulation frequency as a feature. Details
about the rate at which the speaker says words and some other stylistic attributes are
included in the modulation frequency.
Speech intelligibility is measured using less than 20 Hz modulation frequencies.
To achieve the highest efficiency with this function, a temporal window of 300 ms
was used with less than 20 Hz modulation frequencies. Instead of spectrogram mag-
nitudes, discrete cosine transform (DCT) can be applied to temporal trajectories for
the reduction of dimensionality of the spectro-­temporal features. DCT has an advan-
tage over DFT in that it can minimize dimensionality while maintaining the relative
phases of feature vectors, allowing phonetic- and speaker-­specific information to be
contained. Instead of using amplitude-­based approaches, frequency modulation can
 Factors responsible and phases of speaker recognition system  195

be used to improve the unit. To separate speech signals into sub-­band signals, band-­
pass filter bank can be used. Then, using dominant frequency components like fre-
quency centroids, extraction of formant frequency features is done. Center and pole
frequency difference in the sub-­band can then be used as a frequency modulation-­
based function.

9.2.2.4 Prosodic features
Since prosody is such an important aspect of speech interpretation, it is critical to
use prosodic features for the improvement of speech processing. In general, pro-
sodic features are paired with other acoustic features for use in recognition sys-
tems; however, there are some drawbacks, such as the range of prosodic features,
and therefore framework handling segmental features cannot accommodate these
features. These features, which include pause length, syllable stress, speaking rate,
pitch or tempo, intonation patterns, and energy distribution, are referred to as supra-­
segmental features.
Another issue is determining speaker differences through the processing of pro-
sodic data, which can be instantaneous or long-­term. Furthermore, the characteristics
may be dependent
 on aspects that the speaker may alter deliberately. The fundamen-
tal frequency ‍ Fo  ‍is a prosodic function that is commonly used. Combination of
‍Fo‍ (fundamental frequency) and spectral based features has proven to be the most
efficient and reliable in noisy environments. Energy and length, in addition to ‍Fo‍
related features, are much accurate than other prosodic features. ‍Fo‍is now explored
in depth because it is the most critical prosodic function. ‍Fo‍ includes both physi-
ological and learned knowledge and these two information are essential for recog-
nizing a speaker.

9.2.2.5 High-level features
Speakers can also be discriminated against based on the words they use often dur-
ing a conversation. Doddington began doing research in this field in 2001 [78].
To discriminate against the speakers, an idiolect (a speaker’s specific vocabulary)
was used. The idea behind using high-­level features is to turn utterances into the
sequence of tokens and then distinguish various speakers based on the occurrence
of a similar pattern of tokens. These tokens may be words, phonemes, or prosodic
variations such as pitch or energy rises or falls. Figure 9.3 depicts the second and
third phases, namely feature extraction and feature matching.

9.2.3  Feature mapping

Speaker identification/matching is the next, which is done after the features from
speech signals have been extracted. The codebook model, statistic model, and tem-
plate model are the most commonly used speaker recognition models.
Speakers can be identified using a variety of techniques. Major groups of these
techniques:
196  Intelligent network design driven by Big Data analytics

Figure 9.3    Second phase-­feature extraction and final phase-­feature matching

9.2.3.1 Statistical techniques
These techniques include HMM, GMM, UBM (Universal Background Model), and
Vector Quantization (VQ), etc.

9.2.3.2 Soft-computing techniques
Soft-­computing techniques are based on human mind’s cognitive actions. Neural
networks and fuzzy logic are two approaches that are developed by combining vari-
ous methods for solving real-­world problems.

9.2.3.3 Hybrid techniques
These techniques combine mathematical and soft computing techniques to take
advantage of the benefits of both.
 Factors responsible and phases of speaker recognition system  197

Figure 9.4   Human speech production system

9.3 Basics of speech signals

Speech signal is an input for each speech processing technique listed above. As
a result, learning the fundamentals of speech signals is important. The basics of
speech signals are presented here. First and foremost, in order to build good speech
and speaker recognition systems, the human voice development and perception
method and speech signal action must be understood.

9.3.1  Speech production system

Figure  9.4 depicts the speech production system. The diaphragm is displaced to
produce sound, either to sniff into the lungs or to emit air. When air is expelled, it
passes through the windpipe (trachea) and larynx, where vocal cord vibrations can
have an effect. After that, the air travels into the pharyngeal cavity, oral cavity, and
198  Intelligent network design driven by Big Data analytics

nasal cavity [91–99]. The final sound quality is now determined by the arrangement
of articulators: velum (soft palate), tongue, lips, and mandible (jaw).

9.3.2  Speech perception

When compared to machines, humans hear speech sounds differently. Complex
mathematical techniques are not used to analyze sound spectrum. The brain is very
good at separating informative stimuli from distracting noises. These loud noises
may come from variety of sources, such as large engine or a person speaking in
French (if that person does not understand French language).
The human brain ignores almost all types of noise in order to retrieve important
information. Another significant benefit of human speech recognition system is the
capacity of humans to concentrate their attention on a single voice among many
[100–103].
Using a phase mismatch of speech signals reaching the individual ear, the task
of extracting one voice is accomplished. For people who can only listen with one
ear, this intrusion is difficult to eliminate.

9.3.3  Speech signals

Density of medium affects speed at which speech is delivered. The amplitude of
sound signal’s air pressure variation and time is represented by the speech waveform.
This is similar to periodic signal and corresponds to articulated speech sounds
such as vowels and voiced consonants such as “b,” “w,” and “z.” Another form of
speech sound is erratic which is aperiodic and appears to be caused by a random
change in air pressure.
The hissing sounds in s’ and sh’, for example. These speech waveforms corre-
spond to a pure tone and can be used to characterize more complex speech samples.
Speech sounds are defined by a sinusoid. This pure form is very useful when dis-
cussing the properties of speech sounds.

9.3.4  Properties of the sinusoids

When plotted against time, sinusoid is simply a smooth up-­down movement. A sinu-
soid can yield three different types of measurements, all of which together describe
the waveform’s shape. The measurements are broken down into subsections below.

9.3.4.1 Amplitude
Maximum movement of sinusoid above and below the zero-­crossing relation is
called amplitude. The amplitude of a speech signal represents its energy, and hence
its loudness.
The calculation of amplitude can be done in a variety of ways. Amplitude cal-
culation may be performed in a unit of pressure since it is related to degree of air
pressure variation. It is usually expressed in decibels (dB), which is a logarithmic
scale measurement for amplitude relative to normal signal. The dB scale is useful
because it corresponds to how humans experience loudness.
 Factors responsible and phases of speaker recognition system  199

9.3.4.2 Frequency
The number of cycles per second can be used to define frequency. An oscillation
from zero-­crossing reference to a peak, down to a peak below zero-­crossing refer-
ence, and back to the zero-­crossing reference is referred to as a loop.
The unit of frequency measurement is cycle/second. The time duration is the
inverse of the frequency, or the amount of time it takes to complete a loop. Pitch of
a speech sound is believed to change with difference in frequency (though the pitch
is a more complex perceptual quantity).

9.3.4.3 Phase
Phase can be described as the location of sinusoid’s starting point. Sinusoids that
start at maxima have a phase of zero degrees, while those that start at minima have
a phase of 180 degrees.
The phase is extremely difficult to interpret, but relative phase differences
between two signals are much easier to detect. In fact, it is the foundation of human
binaural hearing, as the human brain deduces the location of source of a speech
sound based on phase differences heard in both the ears.

9.3.5  Windowing signals

Speech signals are studied over a brief period of time since signals can be considered
stationary during this time period, requiring the speech signal to be separated into
small chunks. As a result, before study, speech signals are windowed or cut out as a
smaller section.
Signal is multiplied by a window function with an amplitude of one in area of
interest and zero elsewhere, referred to as windowing. As a result, only the win-
dowed signal is kept, while the rest is discarded. Rectangular window’s disadvan-
tage is abrupt shifts at the edges, as well as the distortion created by this in the signal
to be analyzed.
A hamming window, which is relatively smooth window shape, is used to
reduce distortion. It is beneficial to use hamming window, particularly when using
frequency domain techniques. A rectangular window is useful for number of time-­
domain techniques. In order to calculate a broad signal, successive windows must
be used.

9.3.6  Zero-crossing rate

The zero-­crossing rate (ZCR) is the number of zero-­crossings per second, or rate at
which signal crosses midline in one second. This determines primary frequency of
signals, i.e., frequency with greatest amplitude. When RMS and ZCR are combined,
they can easily distinguish between voiced and unvoiced speech.
Unvoiced speech signals have a high zero-­crossing rate, whereas voiced speech
signals have a high RMS, so the properly weighted number of ZCR and RMS would
be high for voiced signals and low for unvoiced signals.
200  Intelligent network design driven by Big Data analytics

Figure 9.5    Autocorrelation of signal

9.3.7  Autocorrelation
Autocorrelation can be used to measure the pitch of voice. This approach is based on
evaluating the relationship between speech signal and its delayed counterpart. When
the delay is a pitch time, the phase of speech and its retarded variant will change, i.e.,
if one signal increases, another will rise as well, and vice versa.
When the delay is half the pitch time, there will be uncorrelated and out-­of-­
phase signals, which means that if one signal rises, others drop, and vice versa. The
autocorrelation curve shown in Figure 9.5 is obtained by plotting the degree of cor-
relation versus lag between signal and delayed version of same signal.
A peak can be seen in plot at a point that relates to the lag of pitch duration. As
a result, autocorrelation can be used to determine a sound’s pitch.

9.4 Features of speech signals

The following characteristics of the speech signal are discussed:

• Physical features
• Perceptual features
• Signal features

9.4.1  Physical features

Following are some of these characteristics:
 Factors responsible and phases of speaker recognition system  201

• Power
• Fundamental frequency
• Spectral features
• Duration of sound

9.4.1.1 Power
It is directly proportional to amplitude in speech signal and can be described as
work done per second. The higher the power present in the signal, the louder the
speech tone. Power calculation may also be used to determine presence of silence in
a speech signal as well as the dynamic range.
Windowing the speech signal, squaring the sample, and obtaining mean after-
ward may be used to evaluate the strength in speech signal for short period of time.
The energy present in a specific frequency band may be used to detect the edges
of speech signal. Applying a hard energy threshold to differentiate between frames
with no signal and frames with low energy, such as the edges of fade, cannot work.

9.4.1.2 Fundamental frequency
The signals that are either periodic or pseudo-­periodic which have a validated funda-
mental frequency, which is denoted   by‍fo.‍ Periodic
  signals repeat themselves indefi-
nitely with a period of ‍ ‍, i.e., ‍ w t +  = w t ‍and ‍fo =  1
 ‍for
 a maximum ‍
    ‍. The
pseudo-­periodic signal nearly repeats itself is represent by ‍ w t +  = w t + " .‍
From one period to the next, the signal varies slightly, but still, ‍fo =  1‍is valid,
where ‍ "‍corresponds to tolerance value. In this case, the signal is periodic if the
calculated ‍fo‍is constant or roughly constant and true for the remaining signal. ‍fo‍is
used to acquire the speech signal’s edges.
The idea is that the significant variance in ‍fo‍is more likely to occur at the end
of word rather than in the middle.

9.4.1.3 Spectral features
Of the many simple spectral features, bandwidth is one of most significant spectral
features. Spectrum of different frequencies available in speech signals is named as
bandwidth. To differentiate music from speech sounds, bandwidth (a spectral fea-
ture) is used.

9.4.1.4 Duration of sound
The period of sound is simply the length of time the sound lasts. This is used as a
function of particular application based on multimedia database. If the unknown
sound’s duration appears to be similar to reference template, duration matching
techniques may be used.
202  Intelligent network design driven by Big Data analytics

9.4.2  Perceptual features

The aim of extracting perceptual features from a speech signal is to determine which
features humans use to distinguish between speech sounds.
The following are a few of them:

• Pitch
• Prosody

9.4.2.1 Pitch
Since pitch includes so much knowledge about speech signals, it tends to be a bet-
ter perceptual function. It appears to be identical to a physical property known as
frequency.
Frequency, on the other hand, is an absolute and numerical quantity, while pitch
is a subjective and fluid quantity. It is perceptible; nevertheless, no human being can
identify a particular pitch value.

9.4.2.2 Prosody
Prosody is a perceptual characteristic of speech that correlates to changes in pitch
and phoneme duration, as well as significant pauses during a spoken word, which
suggest deeper meaning.
It can also be used to emphasize a single word in a sentence. Prosody research
for classification assumes that speech has already been understood and that prosody
offers additional context. In the absence of understood voice, prosody may be used.

9.4.3  Signal features

Signal features are linked to speech sample’s characteristics. In computer systems, a
speech sample reflects the sound of speech.
These characteristics are statistical and quantitative in nature and are commonly
related to how speech signals have been translated into information that computers
can understand.

9.4.3.1 ZCR-related features
ZCR is a measurement of the number of zero-­crossings per second. The spectral  
content of speech signals is determined by ZCR. The fundamental frequency ‍ fo ‍
was determined using ZCR.
By first using filters to filter out high-­frequency components that can degrade
the measurement, a zero-­crossing rate-­based ‍fo‍detector can be created.
However, the cut-­off frequencies of the filters must be chosen with care, as if
they are not, the filters may delete ‍fo‍ partially in order to remove as many high-­
frequency contents as possible.
 Factors responsible and phases of speaker recognition system  203

Figure 9.6    Stages of speaker localization

A ZCR-­based ‍fo‍detector can be used to detect patterns in zero crossings and to

hypothesize a measure of ‍fo-­‍ based patterns.

9.5 Localization of speaker

The method of calculating a person’s position in room based on sound signal

received by an array of microphones is known as speaker localization.
As shown in Figure 9.6, speaker localization can be done in two stages:

1. Calculate the TDE between each microphone pair and the source.
2. Estimated time delays and microphone distance information are used to deter-
mine the source’s position.

Microphone arrays are arranged in a circular pattern in this method, but the
issue is acoustic environment in which array is mounted. The microphones pick up
not just the speech signals but also the reverberated signals as well as the ambient
noise in this situation.
Meeting rooms with various types of sensors have recently become common.
These are referred to as smart meeting spaces, and they use a microphone array to
record multiperson meetings.
Additionally, the registered data can be used to automatically structure and
index meetings.

9.6 Conclusion

On the basis of extensive literature review, we identified the various factors affecting
the database of speech signal. Further study motivates us to explore the various tech-
niques of feature extraction, and feature mapping techniques to identify the speaker
have been explored based on the extracted features of speech signals.

References

[1] Alcazar V.J.L.L., Maulana A.N.M., Mortega II R.O., Samonte M.J.C.

‘Speech-­
to-­
visual approach e-­
learning systems for the deaf’. 11th
204  Intelligent network design driven by Big Data analytics

International Conference on Computer Science and Education, Nagoya,

Japan; IEEE, 2017. pp. 239–43.
[2] Chen S.-H., Wang J.-F., Chen M.-H, et al. ‘A design of far-­field speaker
localization system using independent component analysis with subspace
speech enhancement’. 11th IEEE International Symposium on Multimedia,
2009. ISM’09; San Diego, CA, IEEE, Hyatt Regency Mission Bay Spa and
Marina, 2009. pp. 558–61.
[3] Duda R.O., Martens W.L. ‘Range dependence of the response of a spherical
head model’. The Journal of the Acoustical Society of America. 1998, vol.
104(5), pp. 3048–58.
[4] Fang B.T. ‘Simple solutions for hyperbolic and related position fixes’. IEEE
Transactions on Aerospace and Electronic Systems. 1990, vol. 26(5), pp.
748–53.
[5] Feng W., Guan N., Li Y., Zhang X., Luo Z. ‘Audio visual speech recognition
with multimodal recurrent neural networks’. International Joint Conference
on Neural Networks; 2017. pp. 681–8.
[6] Fernandez-­Lopez A., Martinez O., Sukno F.M. ‘Towards estimating the
upper bound of visual-­speech recognition: the visual lip-­reading feasibil-
ity database’. 12th IEEE International Conference on Automatic Face and
Gesture Recognition; 2017. pp. 208–15.
[7] Xenaki A., Gerstoft P. ‘Grid-­free compressive beamforming’. The Journal of
the Acoustical Society of America. 2015, vol. 137(4), pp. 1923–35.
[8] Richardson M.H., Formenti, D.L. J. ‘Global curve fitting of frequency
response measurements using the rational fraction polynomial meth-
od’. Proceedings of the Third International Modal Analysis Conference,
Publisher: Society for Experimental Mechanics; 1985. pp. 390–97.
[9] Hahn W.R. ‘Optimum signal processing for passive sonar range and bearing
estimation’. The Journal of the Acoustical Society of America. 1975, vol.
58(1), pp. 201–7.
[10] Lee H. ‘A novel procedure for assessing the accuracy of hyperbolic multilat-
eration systems’. IEEE Transactions on Aerospace and Electronic Systems.
1975, vol. AES-­11(1), pp. 2–15.
[11] Schmidt R.O. ‘A new approach to geometry of range difference location’.
IEEE Transactions on Aerospace and Electronic Systems. 1972, vol. AES-­
8(6), pp. 821–35.
[12] Van Etten J. 'Navigation systems. fundamentals of low- and very-­low-­
frequency hyperbolic techniques 192-­212'. Electrical Communication.
1970, vol. 45(3).
[13] Birchfield S.T., Gangishetty R. ‘Acoustic localization by interaural level
difference’. IEEE International Conference on Acoustics, Speech, and
Signal Processing, Proceedings. (ICASSP’05), Vol. 4, IEEE, Philadelphia
Convention Center; Philadelphia, PA, 19–23 Mar; 2005. pp. iv–1109.
[14] Ho K.C., Sun M. ‘An accurate algebraic closed form solution for energy-­
based source localization’. IEEE Transactions on Audio, Speech and
Language Processing. 2007, vol. 15(8), pp. 2542–50.
 Factors responsible and phases of speaker recognition system  205

[15] Ho K.C., Sun M. ‘Passive source localization using time differences of

arrival and gain ratios of arrival’. IEEE Transactions on Signal Processing.
2008, vol. 56(2), pp. 464–77.
[16] Julián P., Andreou A.G., Riddle L., Shamma S., Goldberg D.H.,
Cauwenberghs G. 'A comparative study of sound localization algorithms for
energy aware sensor'. IEEE. 2004.
[17] Xu Z., Liu N., Sadler B.M. ‘A simple closed form linear source localization
algorithm’. IEEE Military Communications Conference, MILCOM 2007;
IEEE, Gaylord Palms Resort and Convention Center, Orlando, FL, 29–31
Oct; 2007. pp. 1–7.
[18] Brandstein M.S., Silverman H.F. ‘A practical methodology for speech source
localization with microphone arrays’. Computer Speech & Language. 1997,
vol. 11(2), pp. 91–126.
[19] Brandstein M.S., Adcock J.E., Silverman H.F. ‘A closed-­form loca-
tion estimator for use with room environment microphone arrays’. IEEE
Transactions on Speech and Audio Processing. 1997, vol. 5(1), pp. 45–50.
[20] Knapp C., Carter G. ‘The generalized correlation method for estimation of
time delay’. IEEE Transactions on Acoustics, Speech, and Signal Processing.
1976, vol. 24(4), pp. 320–7.
[21] Lleida E., Fernández J., Masgrau E. ‘Robust continuous speech recogni-
tion system based on a microphone array’. Proceedings of the 1998 IEEE
International Conference on Acoustics, Speech and Signal Processing; Vol.
1, IEEE, Washington State Convention and Trade Center, Seattle, WA, 12–
15 May; 1998. pp. 241–4.
[22] Omologo M., Svaizer P. ‘Acoustic event localization using a cross power-­
spectrum phase-­ based technique’. IEEE International Conference on
Acoustics, Speech, and Signal Processing, ICASSP-­94; Vol. 2, IEEE,
Adelaide, South Australia, Australia, 19–22 Apr; 1994. pp. II–273.
[23] Razin S. ‘Explicit (noniterative) Loran solution’. Navigation. 1967, vol.
14(3), pp. 265–9.
[24] Stéphenne A., Champagne B. ‘Cepstral prefiltering for time delay estima-
tion in reverberant environments’. International Conference on Acoustics,
Speech, and Signal Processing, ICASSP-­95, Vol. 5, IEEE, ICASSP 1995;
Detroit, MI, 9–12 May; 1995. pp. 3055–8.
[25] Svaizer P., Matassoni M., Omologo M. ‘Acoustic source location in a
three-­
dimensional space using cross power spectrum phase’. IEEE
International Conference on Acoustics, Speech, and Signal Processing,
ICASSP-­97, Vol. 1, IEEE, ICASSP ’97; Munich, Germany, 21–24 Apr;
1997. pp. 231–4.
[26] Zhang C., Florêncio D., Zhang Z. ‘Why does PHAT work well in low noise,
reverberative environments?’. IEEE International Conference on Acoustics,
Speech and Signal Processing, ICASSP 2008, IEEE, ICASSP 2008; Las
Vegas, NV, 31 Mar, 4 Apr; 2008. pp. 2565–8.
[27] Wang H., Chu P. ‘Voice source localization for automatic camera pointing
system in videoconferencing’. IEEE International Conference on Acoustics,
206  Intelligent network design driven by Big Data analytics

Speech, and Signal Processing, ICASSP-­97; Vol. 1, IEEE, Munich, Germany,

21–24 Apr 1997; 1997. pp. 187–90.
[28] Sheng X., Hu Y-­H. ‘Maximum likelihood multiple-­source localization us-
ing acoustic energy measurements with wireless sensor networks’. IEEE
Transactions on Signal Processing. 2005, vol. 53(1), pp. 44–53.
[29] Blatt D., Hero A.O. ‘Energy-­based sensor network source localization via
projection onto convex sets’. IEEE Transactions on Signal Processing.
2006, vol. 54(9), pp. 3614–19.
[30] Cui W., Cao Z., Wei J. ‘Dual-­microphone source location method in 2-­d
space’. IEEE International Conference on Acoustics, Speech and Signal
Processing, ICASSP 2006; Vol. 4, IEEE, Toulouse Pierre Baudis Congress
Center Toulouse, France, 14–19 May; 2006. pp. IV–IV.
[31] Sen S., Nehorai A. ‘Performance analysis of 3-­D direction estimation based
on head-­related transfer function’. IEEE Transactions on Audio, Speech, and
Language Processing. 2009, vol. 17(4), pp. 607–13.
[32] Kulaib A., Al-­Mualla M., Vernon D. ‘2D binaural sound localization for
Urban search and rescue robotic’. The 12th International Conference on
Climbing and Walking Robots; Istanbul, Turkey; 2009. pp. 9–11.
[33] Carter G. ‘Time delay estimation for passive sonar signal processing’. IEEE
Transactions on Acoustics, Speech, and Signal Processing. 1981, vol. 29(3),
pp. 463–70.
[34] Foy W. ‘Position-­location solutions by Taylor-­series estimation’. IEEE
Transactions on Aerospace and Electronic Systems. 1976, vol. AES-­12(2),
pp. 187–94.
[35] Abel J.S., Smith J.O. ‘Source range and depth estimation from multipath
range difference measurements’. IEEE Transactions on Acoustics, Speech,
and Signal Processing. 1989, vol. 37(8), pp. 1157–65.
[36] Eisenhart L.P. A treatise on the differential geometry of curves and surfaces.
26. Ginn, Paderborn dSPACE GmbH Rathenaustraße; 1909.
[37] Sommerville D.M.Y. The Elements of Non-­Euclidean Geometry. 288. North
Chelmsford, MA: Dover Books on Mathematics Courier Corporation;
2012.
[38] Haris B.C., Sinha R. ‘Exploring sparse representation classification for
speaker verification in realistic environment’. Centenary Conference-­
Electrical Engineering; Bangalore: Indian Institute of Science; 2011. pp.
1–4.
[39] Ramos D., Gonzalez‐Dominguez J., Gonzalez‐Rodriguez J. ‘High
‐performance session variability compensation in forensic automatic speak-
er recognition’. The Journal of the Acoustical Society of America. 2010, vol.
128(4),2378.
[40] Kenny P., Boulianne G., Ouellet P., Dumouchel P. ‘Speaker and session vari-
ability in GMM-­based speaker verification’. IEEE Transactions on Audio,
Speech and Language Processing. 2007, vol. 15(4), pp. 1448–60.
[41] Vogt R.J., Baker B.J., Sridharan S. ‘Modelling session variability in text
independent speaker verification’. Eurospeech/Inter-­speech: Proceedings of
 Factors responsible and phases of speaker recognition system  207

the 9th European Conference on Speech Communication and Technology;

2005. pp. 3117–20.
[42] Vogt R., Sridharan S. ‘Explicit modelling of session variability for speaker
verification’. Computer Speech & Language. 2008, vol. 22(1), pp. 17–38.
[43] Benzeghiba M., De Mori R., Deroo O., et al. ‘Automatic speech recognition
and speech variability: a review’. Speech Communication. 2007, vol. 49(10-­
11), pp. 763–86.
[44] Pelecanos J., Sridharan S. ‘Feature warping for robust speaker verifica-
tion’. Proceedings of 2001 A Speaker Odyssey: The Speaker Recognition
Workshop. European Speech Communication Association; ISCA, 2001. pp.
213–18.
[45] Reynolds D.A. ‘Channel robust speaker verification via feature mapping’.
IEEE International Conference on Acoustics, Speech, and Signal Processing,
(ICASSP'03); 2003. pp. II–53.
[46] Mignot R., Chardon G., Daudet L. ‘Low frequency interpolation of room
impulse responses using compressed sensing’. IEEE/ACM Transactions on
Audio, Speech, and Language Processing. 2014, vol. 22(1), pp. 205–16.
[47] Ichinof M., Komatsuff N., Jian-­Gangfff W., Yunffj Y.W. ‘Speaker gen-
der recognition using score level fusion by adaboost’. 11th International
Conference on Control Automation Robotics & Vision; 2010. pp. 648–53.
[48] Wayman J.L. ‘Large-­scale civilian biometric systems-­issues and feasibility’.
Proceedings of Card Tech/Secur Tech ID; 1997. p. 732.
[49] Teunen R., Shahshahani B., Heck L. ‘A model-­based transformational ap-
proach to robust speaker recognition’. Sixth International Conference on
Spoken Language Processing; 2000.
[50] Yin S.-C., Rose R., Kenny P. ‘A joint factor analysis approach to progres-
sive model adaptation in text-­independent speaker verification’. IEEE
Transactions on Audio, Speech and Language Processing. 2007, vol. 15(7),
pp. 1999–2010.
[51] Adami A.G., Mihaescu R., Reynolds D.A., Godfrey J.J. ‘Modeling pro-
sodic dynamics for speaker recognition’. IEEE International Conference on
Acoustics, Speech, and Signal Processing, Proceedings (ICASSP’03), Hong
Kong, China; 2003. pp. IV–788.
[52] Rao K.S., Maity S., Reddy V.R. ‘Pitch synchronous and glottal closure based
speech analysis for language recognition’. International Journal of Speech
Technology. 2013, vol. 16(4), pp. 413–30.
[53] Bhattacharjee U., Sarmah K. ‘A multilingual speech database for speaker rec-
ognition’. IEEE International Conference on Signal Processing, Computing
and Control, Publisher: IEEE, Solan, India; 2012. pp. 1–5.
[54] Pradhan G., Prasanna S.M. ‘Significance of speaker information in wide-
band speech’. National Conference on Communications (NCC), Publisher:
IEEE, Bangalore, India, DOI: 10.1109/NCC.2011.5734710; 2011. pp. 1–5.
[55] Algazi V.R. ‘Physical and filter pinna models based on anthropometry’.
Proceedings of 122nd Convention Audio Engineering Society, Presented at
the 122nd Convention2007 May 5–8 Vienna, Austria; 2007. pp. 718–37.
208  Intelligent network design driven by Big Data analytics

[56] Batteau D.W. ‘The role of the Pinna in human localization’. Proceedings
of the Royal Society of London. Series B, Biological Sciences. 1967, vol.
168(1011), pp. 158–80.
[57] Roffler S.K., Butler R.A. ‘Factors that influence the localization of sound in
the vertical plane’. The Journal of the Acoustical Society of America. 1968,
vol. 43(6), pp. 1255–9.
[58] Oldfield S.R., Parker S.P. ‘Acuity of sound localisation: a topography of
auditory space. II. Pinna cues absent’. Perception. 1984, vol. 13(5), pp.
601–17.
[59] Chenghui G., Heming Z., Zhi T. 'Speaker identification of whispered speech
with perceptible mood'. Journal of Multimedia. 2014, vol. 9(4), pp. 553–61.
[60] Martin J.A., Kung H.-C., Mathews T.J., et al. ‘Annual summary of vital
statistics: 2006’. Pediatrics. 2008, vol. 121(4), pp. 788–801.
[61] Hofman P.M., Van Riswick J.G., Van Opstal A.J. ‘Relearning sound locali-
zation with new ears’. Nature Neuroscience. 1998, vol. 1(5), pp. 417–21.
[62] Brown C.P. Modeling the elevation characteristics of the head related im-
pulse response. [Master’s Theses and Graduate Research]. San Jose State
University Scholar Works; 1996.
[63] Campbell J.P., Shen W., Campbell W.M., Schwartz R., Bonastre J.-F., Matrouf
D. ‘Forensic speaker recognition’. IEEE Signal Processing Magazine. 2009,
vol. 26(2), pp. 95–103.
[64] Hwang S., Park Y., Park Y. ‘Sound source localization using HRTF data-
base’. Proceedings of International Conference on Control, Automation,
and Systems (ICCAS2005), KINTEX Gyeonggi-­Do; Korea (South), 2–5
June; 2005. pp. 751–5.
[65] Ortega-­Garcia J., Gonzalez-­Rodriguez J., Marrero-­Aguiar V. ‘AHUMADA:
a large speech corpus in Spanish for speaker characterization and identifica-
tion’. Speech Communication. 2000, vol. 31(2-­3), pp. 255–64.
[66] Madden S.C., Galliano F., Jones A.P., Sauvage M. ‘ISM properties in
low-­metallicity environments-­I. mid-­infrared spectra of dwarf galaxies’.
Astronomy & Astrophysics. 2006, vol. 446(3), pp. 877–96.
[67] Gravier G., Adda G., Paulson N., Carré M., Giraudel A., Galibert O. ‘The
ETAPE corpus for the evaluation of speech-­based TV content processing in
the French language’. 2012.
[68] Dhall A., Goecke R., Joshi J., Wagner M., Gedeon T. ‘Emotion recogni-
tion in the wild challenge’. Proceedings of the 15th ACM on International
Conference on Multimodal Interaction; 2013. pp. 509–16.
[69] Galibert O., Kahn J. ‘The first official repere evaluation’. First Workshop on
Speech, Language and Audio in Multimedia (SLAM 2013), 43-­48; 2013.
[70] Marković B., Jovic̆ić S.T., Galić J., Grozdić Đ. ‘Whispered speech data-
base: design, processing and application’. International Conference on Text,
Speech and Dialogue; Berlin, Heidelberg: Springer; 2013. pp. 591–8.
[71] Segundo E.S., Alves H., Trinidad M.F. ‘Civil corpus: voice quality for
speaker forensic comparison’. Procedia - Social and Behavioral Sciences.
2013, vol. 95(2), pp. 587–93.
 Factors responsible and phases of speaker recognition system  209

[72] Vloed D.V., Bouten J., van Leeuwen D. ‘NFI-­FRITS: A forensic speaker rec-
ognition database and some first experiments’. The Speaker and Language
Recognition Workshop; Joensuu, Finland, 16–19 Jun 2014; 2014.
[73] Ajili M., Bonastre J.F., Kahn J., Rossato S., Bernard G. ‘Fabiole, a
speech database for forensic speaker comparison’. Proceedings of the
Tenth International Conference on Language Resources and Evaluation
(LREC'16); 2016. pp. 726–33.
[74] Harrington J., Cassidy S. Techniques in Speech Acoustics. Dordrecht:
Kluwer Academic Publishers; 1999.
[75] Smith J., Abel J. ‘Closed-­form least-­squares source location estimation from
range-­difference measurements’. IEEE Transactions on Acoustics, Speech,
and Signal Processing. 1987, vol. 35(12), pp. 1661–9.
[76] Besacier L., Bonastre J.F., Fredouille C. ‘Localization and selec-
tion of speaker-­specific information with statistical modeling’. Speech
Communication. 2000, vol. 31(2-­3), pp. 89–106.
[77] Kinnunen T., Li H. ‘An overview of text-­independent speaker recognition:
from features to supervectors’. Speech Communication. 2010, vol. 52(1), pp.
12–40.
[78] Doddington G. ‘Speaker recognition based on idiolectal differences be-
tween speakers’. Proceedings of Seventh European Conference on Speech
Communication and Technology (Eurospeech 2001), Eurospeech 2001 -
Scandinavia; 2001. pp. 2521–24. Available from https://www.isca-speech.​
org/archive_v0/archive_papers/eurospeech_2001/e01_2521.pdf
[79] Haidar M., Kumar S. 'Smart healthcare system for biomedical and health care
applications using aadhaar and blockchain'. Presented at 5th International
Conference on Information Systems and Computer Networks, ISCON 2021,
Publisher: IEEE; Mathura, India.
[80] Kumar S., Cengiz K., Vimal S., Suresh A., et al. ‘Energy efficient resource
migration based load balance mechanism for high traffic applications IoT’.
Wireless Personal Communications. 2021, vol. 10(3).
[81] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient
multichannel MAC protocol for high traffic applications in heterogene-
ous wireless sensor networks’. Recent Advances in Electrical & Electronic
Engineering. 2017, vol. 10(3), pp. 223–32.
[82] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient
clustering and next hop knowledge based routing in multiple heterogene-
ous wireless sensor networks’. International Journal of Grid and High
Performance Computing. 2017, vol. 9(2), pp. 1–20.
[83] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering (IJITEE). 2019, vol. 8(6).
[84] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
210  Intelligent network design driven by Big Data analytics

[85] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware distrib-
uted protocol for heterogeneous wireless sensor network’. International Journal
of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[86] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality moni-
toring using machine learning algorithm’ in Reddy A.N.R., Marla D.,
Favorskaya M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and
Energy Sustainability. Smart Innovation, Systems and Technologies. 265.
Singapore: Springer; 2022.
[87] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and iot for
smart road traffic management system’. Proceedings of the 2020 IEEE India
Council International Subsections Conference, INDISCON 2020, Publisher:
IEEE, Visakhapatnam, India; 2020. pp. 289–96.
[88] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementa-
tion of fault tolerance technique for internet of things (iot)’. Proceedings
of the 12th International Conference on Computational Intelligence and
Communication Networks, CICN 2020, Publisher: IEEE, Bhimtal, India;
2020. pp. 154–59.
[89] Singh P., Bansal A., Kumar S. ‘Performance analysis of various informa-
tion platforms for recognizing the quality of indian roads’. Proceedings of
the Confluence 2020 - 10th International Conference on Cloud Computing,
Data Science and Engineering, Publihser: IEEE, Noida India; 2020. pp.
63–76.
[90] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. Proceedings of the
Confluence 2020-­10th International Conference on Cloud Computing, Data
Science and Engineering, Publisher: IEEE, Noida, India; 2020. pp. 63–76.
[91] Reghu S., Kumar S. ‘Development of robust infrastructure in networking
to survive a disaster’. 4th International Conference on Information Systems
and Computer Networks, ISCON 2019, Publisher: IEEE, Mathura, India;
2019. pp. 250–55.
[92] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
Proceedings of the International Conference on Computational Intelligence
and Communication Networks, CICN 2015, Publisher: IEEE, Jabalpur,
India; 2016. pp. 79–84.
[93] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maxi-
mization approach to MAC layer channel access and forwarding’.
Progress in Electromagnetics Research Symposium, PIERS 2015, The
Electromagnetics Academy, 6-­9 July 2015, Prague, Czech Republic;
2015. pp. 2363–67.
[94] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wire-
less sensor network’. Proceedings of the 5th International Conference on
Communication Systems and Network Technologies, CSNT 2015; Gwalior,
India, IEEE, 2015. pp. 194–200.
 Factors responsible and phases of speaker recognition system  211

[95] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distributed

localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT), IEEE; Ghaziabad, India, 2014.
[96] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems.; Singapore: Springer; 2020.
[97] Kumar S., Trivedi M.C., Ranjan P., Punhani A. Evolution of Software-­
Defined Networking Foundations for IoT and 5G Mobile Networks. Hershey,
PA: IGI Publisher; 2020. p. 350.
[98] Sampathkumar A., Rastogi R., Arukonda S., Shankar A., Kautish S., Sivaram
M. ‘An efficient hybrid methodology for detection of cancer-­causing gene
using CSC for micro array data’. Journal of Ambient Intelligence and
Humanized Computing. 2020, vol. 11(11), pp. 4743–51.
[99] Nie X., Fan T., Wang B., Li Z., Shankar A., Manickam A. ‘Big data analyt-
ics and IoT in operation safety management in under water management’.
Computer Communications. 2020, vol. 154(1), pp. 188–96.
[100] Shankar A., Jaisankar N., Khan M.S., Patan R., Balamurugan B. ‘Hybrid
model for security‐aware cluster head selection in wireless sensor networks’.
IET Wireless Sensor Systems. 2019, vol. 9(2), pp. 68–76.
[101] Shankar A., Pandiaraja P., Sumathi K., Stephan T., Sharma P. ‘Privacy pre-
serving E-­voting cloud system based on ID based encryption’. Peer-­to-­Peer
Networking and Applications. 2021, vol. 14(4), pp. 2399–409.
[102] Kumar A., Abhishek K., Nerurkar P., Ghalib M.R., Shankar A., Cheng X.
'Secure smart contracts for cloud‐based manufacturing using ethereum
blockchain'. Transactions on Emerging Telecommunications Technologies.
2022, vol. 33(4), e4129.
[103] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence
unified with big data analytics, internet of things and cloud computing
technologies’. 5th International Conference on Information Systems and
Computer Networks (ISCON); 2021. pp. 1–6.
This page intentionally left blank
 Chapter 10
IoT-­based water quality assessment using fuzzy
logic controller
A. Hendry 1, K. Ghousiya Begum 1, A. Hariharan 1, and
K. J. Aravind 1

Water is an essential resource that we use in our daily life. The standard of the water
quality must be observed in real time to make sure that we obtain a secured and clean
supply of water to our residential areas. A water quality-­monitoring and decision-­
making system (WQMDMS) is implemented for this purpose based on Internet of
Things (IoT) and fuzzy logic to decide the usage of water (drinking or tap water)
in a common water tank system. The physical and chemical properties of data are
obtained through continuous monitoring of sensors. The work describes in detail the
design of a fuzzy logic controller (FLC) for a water quality measurement system,
to determine the quality of water by decision-­making, and accordingly, the usage of
water is decided. The WQMDM system measures the physico-­chemical characteris-
tics of water like pH, turbidity, and temperature by the use of corresponding analog
and digital sensors. The values of the parameters obtained are used to detect the
presence of water contaminants and accordingly, the quality of water is determined.
The measurements from the sensor are handled and processed by ESP32, and these
refined values follow the rules determined by the fuzzy inference system (FIS). The
output highlights the water quality that is categorized as very poor, poor, average,
and good. The usage of the water will be determined by the results obtained using
the FLC and as per the percentage of water quality, the water is decided as drinking
water or tap water.

10.1 Introduction

To understand the quality of water, we should know the chemical, physical, and
natural features of the water based on the norms of its operation. It is not easy to
say that “water is good” or “water is bad.” Water quality is generally determined in

1
School of EEE, SASTRA Deemed University, Thirumalaisamudram, Thanjavur, Tamil Nadu, India
214  Intelligent network design driven by Big Data analytics

relation to the usage of water. ESP32 microcontrollers are used in IoT-­based water
quality monitoring systems. The sensing devices like pH, turbidity, and DS18B20
temperature sensors are interfaced with ESP32. The ESP32 receives the data from
sensors and updates them in the cloud platform ThingSpeak with the help of the
Internet via Wi-­Fi. The live data from ThingSpeak will be fed into MATLAB® as the
input for the fuzzy logic system.
Fuzzy logic is a computational approach to representing the vagueness or the
uncertainty rather than mentioning the “good or bad” (the integer values 1 or 0)
boolean sense on which ultra-­modern computers are based. Fuzzy logic is an intro-
ductory control system that depends on the state of the input and affair depends on
the rate of change of that state.
In the work done by Unnikrishna Menon et al. [1], a system is described for
the quality assessment and monitoring of the river water based on wireless sensor
networks (WSN) that support both remote and continuous monitoring of the quality
levels of water in India. In their work, they used only a pH sensor and tested the dif-
ferent conditions of pH like lemon juice, rainwater, and drinking water. They used
Zigbee technology to transmit the data from sensing devices to the cluster head.
Bokingkito and Caparida [2] proposed a water quality assessment system for
finding out the decrease in aquaculture (fisheries) production. He and Zhang [3]
presented a work based on wireless water monitoring networks and remote data cen-
ters and suggested using the CC2430 microprocessor as a core hardware platform.
The WSN is constructed on the Zigbee communication module. The WSN scans
the parameter and uses the GPRS DTU with the TCP/IP protocol to send the live
readings to the Internet. Pande et al. [4] presented a manuscript to assess the quality
of drinking water for housing society. The proposed system collects the parameters
like temperature, turbidity, level, and pH to measure the quality of water samples.
They suggested the use of ESP8266 Wemos d1 mini and Raspberry pi for simple,
faster, efficient, real-­time, monitoring of data. Sarwar et al. [5] presented a study on
the designed fire detection and warning systems for buildings based on a fuzzy logic
theory and carried out the simulation tasks in the MATLAB Fuzzy Logic toolbox.
They used Arduino Uno R3 and fuzzy logic to predict only the true incidents of fire
with the data obtained from temperature, humidity, and flame sensor. They proposed
a control mechanism that can activate water showers when it detects fire. Lambrou
et al. [6] worked on the design of monitoring water quality at consumer sites using
the optical and electrochemical sensors installed in the pipelines. They proposed a
system with a PIC32 MCU board that gathers water quality parameters from sensors
and sends the data to the ARM platform and to the Internet that stores data and send
email or message to the notification node through Zigbee. Using event detection
algorithms, the alarms were activated to detect the water quality standards.
Faruq et al. [7] designed and implemented a cost-­effective, simple water quality
assessment system with calibrated sensors for measuring parameters like tempera-
ture, turbidity, and pH, which will be shown on the LCD monitor. They just detected
the water quality without using IoT. Vigueras-Velázquez et al. [8] discussed a work
to evaluate the freshwater quality in farming tanks to grow whitefish (Chirostoma
estor water quality) and used sensors to measure dissolved oxygen (DO), pH,
 IoT-­based water quality assessment using fuzzy logic controller  215

Figure 10.1   Working model

temperature, non-­ionized ammonia, and total ammonia. The main aim is to maintain
the ideal conditions for the sustained growth of fish. Better aquaculture water assess-
ment was made probable by the implementation of weighted FIS. Pasika and Gandla
[9] also proposed a smart water quality assessment system using Arduino mega with
ultrasonic, pH, turbidity, temperature sensors, and Wi-­Fi ESP8266 node MCU with
the cloud platform. Baghavan and Saranya [10] proposed a sensor network with AI
to identify the pollutants in water so that the water can be subjected to a purification
process. Kothari et al. [11] developed a system to test the rainwater, tap water, well
water, and purified RO water using sensors, viz., for measuring temperature, TDS,
pH, and DO, along with Arduino mega2560 and GSM module. Chowdury et al. [12]
also discussed how the WSN along with a microcontroller is used for processing and

Figure 10.2   Arduino IDE and monitor window

216  Intelligent network design driven by Big Data analytics

Figure 10.3   E-­201C pH sensor

establishing the inter- and intra-­node communication among sensors for pollutant
monitoring of water for Bangladeshi populations. They were able to acquire the real-­
time data and were able to access using remote monitoring and IoT. The collected
data were displayed on a PC through an expert system and DL neural network mod-
els, in comparison with the typical values that generate computerized alert messages
if the obtained value is above the threshold limit [13–20].
Based on the motivation from the previous studies, the proposed system of
monitoring the water quality and decision-­making has been implemented in real
time for consumer application using only ESP32 with pH, temperature, and turbidity
sensors for processing and monitoring, and the fuzzy logic system is implemented
for decision-­making. It does not require any data centers as we are storing and dis-
playing the measured water quality parameters in the cloud platform. This proposed
system is economical and cost effective, and the prediction will be more accurate.

Figure 10.4   Analog turbidity sensor

 IoT-­based water quality assessment using fuzzy logic controller  217

Figure 10.5   DS18B20 temperature sensor

We can measure the water quality of water tank systems that are installed in multiple
locations and also get the live data in the cloud system [21–36].

10.2 Experimental procedures

The proposed system incorporates an ESP32 microcontroller and sensors like

E-­201C for measuring pH value, analog turbidity sensor, and DS180B20 digital

Figure 10.6   ESP32 microcontroller

218  Intelligent network design driven by Big Data analytics

Figure 10.7   Hardware circuit design

sensor to sense the temperature of the water, as shown in Figure 10.1. The hardware
is configured, and the C language program code is scripted in Arduino IDE to obtain
the anticipated format of the sensor data. The sensor parameters, namely, temper-
ature, turbidity, and pH values can be seen in Arduino serial monitor, as shown
in Figure 10.2 which is sent to Thinspeak, and the respective data are obtained in
MATLAB via Thing speak. In MATLAB, using fuzzy logic rules, the degree of
quality, and clearness of the water are determined.
Sensors and microcontroller
pH sensor (E-­201C): The pH of any solution is a quality metric that represents
the acidity or basicity. The pH scale is a log scale of hydrogen ions in solution
(range of 0–14 and a neutral point of 7). The values above seven determine a base or
alkaline solution, and values below seven indicate an acidic solution. It operates on
a 5-­V power supply and is presented in Figure 10.2. Its response time is <1 min and
internal resistance is less than or equal to 250 MΏ. For drinking, water pH should
be between 6.5 and 8.5. The sensor needs to be calibrated where it will display the
voltage and pH value.

Figure 10.8   Fuzzy inference system

 IoT-­based water quality assessment using fuzzy logic controller  219

Figure 10.9   Fuzzy logic designer

Turbidity sensor: The turbidity sensor measures the opaqueness or cloudi-

ness in the water. It measures the transmittance rate and scattering rate of visible
light to identify suspended particles in water like clay, slit, microorganisms, and
organic matter. The operating voltage and current are 5 V DC and 40 mA. The
turbidity changes with the quantity of total suspended solids in water. It also gives
an analog output of 0–4.5 V. For drinking water, the turbidity value should be
between 0 and 5 NTU. The response time is <500 ms. The module size is 38.6
mm × 22.1 mm.
Temperature sensor (DS18B20): The temperature of the water specifies the
hotness or coldness of the water. The DS18B20 temperature sensor whose usable
temperature ranges from –55 to +125°C. This is a digital temperature sensor that
provides an accurate reading. It utilizes one-­wire interface and has 9–12-­bit select-
able resolution.
ESP32: The proposed system uses this microcontroller that is the brain of the
proposed system. ESP32 is programmable hardware with 36 pins and designed to
control the circuit logically and it can be programmed by using Arduino IDE. It has
inbuilt Wi-­Fi and Bluetooth. Its operating voltage ranges from 2.3 to 3.6 V, and
its operating current is around 80 mA. It has a flash memory of 4 MB and SRAM
memory of 512 kB. Its clock frequency ranges from 80 to 240 MHz and its data rate
is 54 Mbps.
The components and sensors used are shown in Figures 10.2–10.6 .
220  Intelligent network design driven by Big Data analytics

Figure 10.10   Membership function for temperature

10.3 Working

The suggested system adopts three sensing devices (temperature, pH, and turbid-
ity), ESP32, and the ThingSpeak platform. Only one microcontroller with inbuilt
Wi-­Fi and Bluetooth module is used. Obtaining the input values from the sensors
and determining the degree of membership level for each input value from sensors
using fuzzy expressions are the first stage in utilizing fuzzy logic rules to evaluate
the water condition (also known as membership function). The sensors (pH, turbid-
ity, and DS18B20 temperature sensor) that are interfaced with the ESP32 micro-
controller are processed using the program in Arduino IDE from which the sensor
readings will be sent to ThingSpeak. The live readings from ThingSpeak will be
fed into the FLC in MATLAB. Figure 10.7 shows the hardware circuit connections.
Fuzzy logic designer
Figure 10.8 depicts a block schematic of the proposed system that uses the
fuzzy-­based decision-­making system. These steps of the proposed fuzzy logic
decision-­making system comprise (a) initialization of linguistic variables, member-
ship functions, and construction of rules, (b) fuzzification where the crisp values of
input data are converted to fuzzy values using membership functions, (c) evaluation
of knowledge-­based rules and combining the results of each rule, and (d) defuzzifi-
cation where the output values are transformed to non-­fuzzy values.
The hardware and software components like ESP32, the three sensors (tempera-
ture, pH, and turbidity), Matlab, and ThingSpeak make up the whole system. The
 IoT-­based water quality assessment using fuzzy logic controller  221

Figure 10.11   Membership function for pH

diagrammatic description of fuzzy logic is shown in Figure 10.8, which indicates

the initial stage in applying fuzzy rules to identify the grade of membership func-
tions of each input vector using FIS. After evaluating the water condition, the input
values are allotted to the output vector, and the rules are constructed by defining
the membership functions. The fuzzy input can contain up to three states from a
linguistic standpoint. The specific rules allow the user in determining the conditions
of input variables as low, mid, and high. The goal of this research is to create a
fuzzy logic system that can classify water quality into four levels and show real-­time
data in ThingSpeak. The MATLAB-­modeled fuzzy logic control system is shown
in Figure 10.9. In this figure, the input variables are defined as temperature, pH, and
turbidity. The FIS is implemented using the Mamdani method. Finally, the output
variables are displayed as four levels of quality.
Membership function
Here, the crisp values are represented as fuzzy values with the help of member-
ship functions. The fuzzification process adopted for every input and output param-
eter is discussed in the following.
Membership function of temperature
As shown in Figure 10.10, the three fuzzy (linguistic) variables for input tem-
perature are used and shown as membership functions. They are labeled as low,
medium, and high. The universe of discourse has units in the range of –50º to 100º.
222  Intelligent network design driven by Big Data analytics

Figure 10.12   Member function for turbidity

Figure 10.13   Water quality membership function

 IoT-­based water quality assessment using fuzzy logic controller  223

Figure 10.14   Fuzzy logic rule editor

The low has a range of –50º to 20º, the medium has a range of 20º to 40º, and the
high has a range of 40º–100º. The triangular membership function is used.
Membership function of pH
Figure 10.11 displays three levels of pH, which are acid, neutral, and base. The
unit of pH has a range of 0–14 pH. The acid has a range of 0–6.5 pH, the neutral has
a range of 6.5–7.5 pH, and the base has a range of 7.5–14 pH.
Membership function of turbidity
Figure 10.12 shows three levels of turbidity as low, mid, and high. The unit of
turbidity has a range of 0–200 NTU. The low has a range of 0–5 NTU, the medium
has a range of 5–30 NTU, and the high has a range of 30–200 NTU.
Membership function of output water quality
The linguistic variables of the output are determined into four categories (very
poor, poor, average, and good). The quality of water is determined by the range of
variables as very poor from 0% to 30%, poor as 30% to 60%, average as 60% to
80%, and good as 80% to 100%, which are mentioned in Figure 10.13.
Fuzzy rule editor for water quality
Figure 10.14 represents the MATLAB rule editor, where rule sets for fuzzy
water quality are implemented. The pH is represented by the first box, the turbidity
variables are represented by the second box on the left, the temperature variables are
represented by the third box on the right, these sides of the box indicate the input
variables, and the level of water quality is represented by the right box.
The results of four different samples of water are displayed as rule viewer
in Figures  10.15–10.18. The first three columns represent the pH, turbidity, and
224  Intelligent network design driven by Big Data analytics

Figure 10.15   Rule viewer for water sample 1

Figure 10.16   Rule viewer for water sample 2

temperature, and the last column indicates the water quality in percentage.
Figure 10.15 displays the first water sample quality of percentage 70 as pH value is
7.05, turbidity value is 24.1 NTU, and the temperature is 34.2°C, which indicates
that it can be used as tap water for cleaning and washing. The second water sample is
taken whose pH is 5.2, turbidity value is 110 NTU, the temperature value is 28.9°C,
and the water quality percentage is 15, as shown in Figure 10.16 and cannot be used
as drinking water. The third water sample is taken whose pH is 6.85, turbidity value
 IoT-­based water quality assessment using fuzzy logic controller  225

Figure 10.17   Rule viewer for water sample 3

is 2.19 NTU, the temperature value is 27.8°C, and the water quality percentage is 90,
as shown in Figure 10.17. This implies that the quality of water is good and can be
used as drinking water. The final water sample is taken whose pH is 8.18, turbidity
value is 81 NTU, the temperature value is 46.5°C, and the water quality percentage is
50, as shown in Figure 10.18, which can be used for cleaning purposes. Figure 10.19
displays the surface viewer of the water quality decision-­making system.
Defuzzification

Figure 10.18   Rule viewer for water sample 4

226  Intelligent network design driven by Big Data analytics

Figure 10.19  Surface viewer for water quality decision-­making system: (a)

pH, turbidity versus output and (b) turbidity, temperature versus
output

The whole monitoring and decision-­making of the water quality system are now
complete as inference and defuzzification are built-­in functions that are executed
by MATLAB. Therefore, defuzzification is the final step in implementing the FLC
where the output is expressed as if condition statements and stored in the knowledge-­
based system database. Here, fuzzification of the scalar values, application of rules,
generation of fuzzy output, and conversion to scalar quantity take place. The com-
monly issued defuzzification methods are centroid and weighted average methods.
The centroid method is used here as it provides accurate and efficient [5]. After the
defuzzification process, the water quality percentage is displayed.
Figure 10.20 shows the hardware connections of the sensors (pH, turbidity, and
DS18B20 temperature sensor) with the ESP32 microcontroller and interfaced with
the Arduino IDE; from there, the sensor readings will be sent to ThingSpeak.

Table 10.1   Decision-­making of water quality system

Sample Parameter Measured value Water quality Decision

%
1 pH 7.104 70 Activate tap water
Turbidity 15.12 NTU motor
Temperature 27.06°C
2 pH 7.05 90 Activate drinking
Turbidity 3.4 NTU water motor
Temperature 27.75 °C
3 pH 6.871 50 Activate tap water
Turbidity 200 NTU motor
Temperature 27.75 °C
 IoT-­based water quality assessment using fuzzy logic controller  227

Figure 10.20   Hardware circuit

10.4 Results and discussions

Several trials were done to evaluate the proposed water monitoring system’s per-
formance. The system’s performance is determined by the delicate nature of the
water. The conducted trials listed below represent the precision of water quality.
Several trials were conducted to evaluate the proposed water quality monitoring
system’s performance, as shown in Tables 10.1 and 10.2. Out of these trials, three
trial instances are visually represented as trials with bad, average, and acceptable
water quality, respectively. The water quality index is indicated in the trials below
from 0% to 100% in terms of the rate of changes in pH, turbidity, and temperature.
ThingSpeak data
ThingSpeak is an open-­source cloud-­based platform for gathering, envisag-
ing, and analyzing live data received from sensors. The ThingSpeak application is

Table 10.2   Table of FLC results

Turbidity Temperature
Water sample pH NTU o
C Water quality % Quality level
1 6.8 3.4 28.8 90 Good
2 6.2 15 30 70 Average
3 7.1 15.3 27.5 70 Average
4 6.9 28 7.25 50 Poor
5 6.3 200 27 30 Very poor
6 7.2 2 31 90 Good
7 8 20 26 40 Poor
8 6 10 40 50 Poor
9 6.85 2.19 27.8 90 Good
10 8.18 81 46.5 50 Poor
11 5.2 110 28.9 15 Very poor
228  Intelligent network design driven by Big Data analytics

Figure 10.21   Sample 1 sensor readings in ThingSpeak

available as a MATLAB library function, and we use it to get real-­time water metrics.
The data from the sensors are displayed in the ThingSpeak interface as shown in
Figures 10.21–10.23) which is then sent to the FLC, and the water quality is calcu-
lated and displayed for three different samples.
Results of MATLAB implementation
The water quality percentage of the three different samples determined using the
FLC designed in MATLAB is displayed in Figures 10.24–10.26.
After the experiments, all the results were obtained, and the data are given in
Tables 10.1 and 10.2. As you can see from the tables, the percentage of water quality
is obtained, ranging from 15% to 90%, which explicates that the suggested system is
working in accordance with the defined rules in the FLC.

10.5 Conclusion

In the proposed system, we have successfully constructed the hardware and soft-
ware tools for WQMDMS using the ESP32 microcontroller, sensors like pH, tur-
bidity, and DS18B20 temperature sensor, MATLAB-­based fuzzy logic control, and
ThingSpeak cloud platform. The circuit operates in error-­free conditions and pro-
vides us with the expected output. Continuous data samples can be obtained using
the present proposed system that will help us to determine the water quality more
accurately. The proposed system is portable to determine the water quality tank sys-
tems across multiple areas. We can also determine the leakages in the pipeline by
IoT-­based water quality assessment using fuzzy logic controller  229

Figure 10.22   Sample 2 sensor readings in ThingSpeak

Figure 10.23   Sample 3 sensor readings in ThingSpeak

230  Intelligent network design driven by Big Data analytics

Figure 10.24   FLC result for sample 1

Figure 10.25   FLC result for sample 2

Figure 10.26   FLC result for sample 3

seeing the variation in water quality levels across the different locations. The future
scope of the system is as follows:

• Controlling the motor system of the water tank to pump water

• To sense and process more water quality parameters
• To develop the water distribution system using pressure and flow sensor
• Getting alert notifications for the water quality levels through the mobile app
• To detect leakages in pipelines.
 IoT-­based water quality assessment using fuzzy logic controller  231

References

[1] Unnikrishna Menon K.A., Divya P., Ramesh M.V. ‘Wireless sensor network
for river water quality monitoring in india’. Third International Conference
on Computing, Communication and Networking Technologies (ICCCNT’12),
IEEE; Coimbatore, India, 2012. pp. 1–7.
[2] Bokingkito P.B., Caparida L.T. ‘Using fuzzy logic for real - time water quality
assessment monitoring system’. Proceedings of the 2018 2nd International
Conference on Automation, Control and Robots (ICACR 2018). Publisher:
Association for Computing Machinery (ACM); Bangkok Thailand, 2018. pp.
21–25.
[3] He D., Zhang L.X. ‘The water quality monitoring system based on WSN’.
2nd International Conference on Consumer Electronics, Communications
and Networks (CECNet), IEEE; Yichang, China, 2012. pp. 3661–64.
[4] Pand A.M., Warhade K.K., Komati R.D. ‘Water quality monitoring system
for water tanks of housing society’. International Journal of Electronics
Engineering Research. 2017, vol. 9(7), pp. 1071–8.
[5] Sarwar B., Bajwa I., Ramzan S., Ramzan B., Kausar M. ‘Design and applica-
tion of fuzzy logic based fire monitoring and warning systems for smart build-
ings’. Symmetry. 2018, vol. 10(11),615.
[6] Lambrou T.P., Anastasiou C.C., Panayiotou C.G., Polycarpou M.M. ‘A low-­
cost sensor network for real-­time monitoring and contamination detection in
drinking water distribution systems’. IEEE Sensors Journal. 2014, vol. 14(8),
pp. 2765–72.
[7] Faruq M.O., Emu I.H., Haque M.N., Dey M., Das N.K., Dey M. ‘Design
and implementation of cost-­effective water quality evaluation system’. IEEE
Region 10 Humanitarian Technology Conference (R10-­ HTC); Publisher:
IEEE, Dhaka, Bangladesh, 2017. pp. 860–63.
[8] Vigueras-­Velázquez M.E., Carbajal-­Hernández J.J., Sánchez-­Fernández L.P.,
Vázquez-­Burgos J.L., Tello-­Ballinas J.A. ‘Weighted fuzzy inference system
for water quality management of Chirostoma estor estor culture’. Aquaculture
Reports. 2020, vol. 18, p. 100487.
[9] Pasika S., Gandla S.T. ‘Smart water quality monitoring system with cost-­
effective using IoT’. Heliyon. 2020, vol. 6(7) e04096.
[10] Bhagavan N.V.S., Saranya P.L. 'Water pollutants monitoring based on internet
of things' in Inorganic pollutants in water. Elsevier; 2020. pp. 371–97.
[11] Kothari N., Shreemali J., Chakrabarti P., Poddar S. 'Design and implemen-
tation of iot sensor based drinking water quality measurement system'.
Materials Today: Proceedings. 2021, vol. 3, pp. 1–10.
[12] Chowdury M.S.U., Emran T.B., Ghosh S., et al. ‘IoT based real-­time river
water quality monitoring system’. Procedia Computer Science. 2019, vol.
155(3), pp. 161–8.
[13] Kumar S., Cengiz K., Trivedi C.M., et al. ‘DEMO enterprise ontology with
a stochastic approach based on partially observable Markov model for data
232  Intelligent network design driven by Big Data analytics

aggregation and communication in intelligent sensor networks’. Wireless

Personal Communication. 2022.
[14] Kumar V., Arablouei R., Cengiz K., Vimal S., Suresh A. ‘Energy efficient re-
source migration based load balance mechanism for high traffic applications
IoT’. Wireless Personal Communications. 2022, vol. 10(3), pp. 1623–44.
[15] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[16] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[17] Punhani A., Faujdar N., Kumar S. 'Design and evaluation of cubic torus
network-­on-­chip architecture'. International Journal of Innovative Technology
and Exploring Engineering, Publisher: Blue Eyes Intelligence Engineering.
2019, vol. 8(6), pp. 1672–76.
[18] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[19] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[20] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring
using machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[21] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and iot for
smart road traffic management system’. IEEE India Council International
Subsections Conference (INDISCON), IEEE; Visakhapatnam, India, 2020.
pp. 289–96.
[22] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation
of fault tolerance technique for internet of things (iot)’. 12th International
Conference on Computational Intelligence and Communication Networks
(CICN), IEEE; Bhimtal, India, IEEE, 2020. pp. 154–59.
[23] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of Indian roads’. 10th International
Conference on Cloud Computing, Data Science & Engineering (Confluence),
IEEE; Noida, India, 29–31 Jan; 2020. pp. 63–76.
[24] Reghu S., Kumar S. ‘Development of robust infrastructure in networking
to survive a disaster. 4th International Conference on Information Systems
and Computer Networks, (ISCON); Mathura, India, 21–22 Nov; 2019. pp.
250–5.
 IoT-­based water quality assessment using fuzzy logic controller  233

[25] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’. 2015
International Conference on Computational Intelligence and Communication
Networks (CICN), IEEE; Jabalpur, India, 2016. pp. 79–84.
[26] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium, Publisher: The Electromagnetics
Academy; Prague, Czech Republic, 2015. pp. 2363–67.
[27] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wireless
sensor network’. Fifth International Conference on Communication Systems
and Network Technologies, IEEE; Gwalior, India, 2015. pp. 194–200.
[28] Kumar S., Ramaswami R., Rao A.L.N. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT), IEEE; Ghaziabad, India, 2014. pp. 350–55.
[29] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based trans-
parent and secure decentralized algorithm’. International Conference on
Intelligent Computing and Smart Communication. Algorithms for Intelligent
Systems, Springer; THDC-­IHET, 2020. pp. 327–36.
[30] Kumar S., Trivedi M.C., Ranjan P. Evolution of Software-­Defined Networking
Foundations for IoT and 5G Mobile Networks. Hershey, PA: IGI Publisher;
2020. p. 350.
[31] Sampathkumar A., Rastogi R., Arukonda S., Shankar A., Kautish S., Sivaram
M. ‘An efficient hybrid methodology for detection of cancer-­causing gene
using CSC for micro array data’. Journal of Ambient Intelligence and
Humanized Computing. 2020, vol. 11, pp. 4743–51.
[32] Nie X., Fan T., Wang B., Li Z., Shankar A., Manickam A. ‘Big data analyt-
ics and IoT in operation safety management in under water management’.
Computer Communications. 2020, vol. 154(1), pp. 188–96.
[33] Shankar A., Jaisankar N., Khan M.S., Patan R., Balamurugan B. ‘Hybrid
model for security‐aware cluster head selection in wireless sensor networks’.
IET Wireless Sensor Systems. 2019, vol. 9(2), pp. 68–76.
[34] Shankar A., Pandiaraja P., Sumathi K., Stephan T., Sharma P. ‘Privacy pre-
serving E-­voting cloud system based on ID based encryption’. Peer-­to-­Peer
Networking and Applications. 2021, vol. 14(4), pp. 2399–409.
[35] Bhardwaj A., Shah S.B.H., Shankar A., Alazab M., Kumar M., Gadekallu
T.R. ‘Penetration testing framework for smart contract blockchain’. Peer-­to-­
Peer Networking and Applications. 2021, vol. 14(5), pp. 2635–50.
[36] Kumar A., Abhishek K., Nerurkar P., Ghalib M.R., Shankar A., Cheng X.
‘Secure smart contracts for cloud-­ based manufacturing using Ethereum
blockchain’. Transactions on Emerging Telecommunications Technologies.
2020, vol. 33(4).
This page intentionally left blank
 Chapter 11
Design and analysis of wireless sensor
network for intelligent transportation and
industry automation
Prabhakar D. Dorge 1, Prasanna M. Palsodkar 2, and Divya
Dandekar 3

This work is based on the wireless sensor networks (WSN), which contain an insuf-
ficient number of device nodes, regularly similarly stated nodes or sensors, and
sensor knots that are associated with all other wireless communications. There are
numerous assumptions or overall possessions of WSNs, and a lot more applica-
tions of WSNs around the creation are presented, making it unbearable to protect all
their application areas. Applications of WSNs span ecological and animal monitor-
ing, factory and manufacturing monitoring, farming monitoring and mechanization,
healthiness monitoring, and many other areas. One of the most characteristics of
WSNs is that they are strongly coupled with their application. In this chapter, WI-­
MAX without wormhole attack is explained, and the related results are explained
with their outputs The NS2 evaluation system is applied to production out of all
imitations.

11.1 Introduction
A wireless network is an all type of processer net that uses wireless information net-
works to wad system nodes. Wireless networks are processer networks that are not
associated by chains irrespective of the kind. The apple of a wireless network allows
originalities to stop the exclusive resources of giving cables into constructions or
as an association connecting different apparatus locations. The foundation of wire-
less systems is the receiver effect, an application to happen on the animal advanced
equal of system construction. Wireless skills are different based on the sizes, for the

1
Department of Electronics and Telecommunication Enginnering, Yeshwantrao Chavan College of
Engineering, Nagpur, India
2
Department of Electronics Engineering, Yeshwantrao Chavan College of Engineering, Nagpur, India
3
Department of Electronics Engineering, Hochschule Breman University, USA
236  Intelligent network design driven by Big Data analytics

most part particularly in just how much bandwidth they offer and how distant to
one side interactive nodes can be real. Additional significant difference comprises
which possible electromagnetic ranges they choose and just how much power they
use. In this part, we deliberate four protuberant wireless technologies: Bluetooth,
Wi-­Fi, Wi-­MAX, and 3G cellular wireless. In the segments, we present them in a
manner from the straight series to the lengthiest range one; for the most part, usu-
ally wireless links are used in these days, which are generally unequal, i.e., together
endpoints are typically kinds of nodes. One endpoint, from time to time called the
base station, generally has no flexibility but has a connection to the Internet or other
systems that join in the conflicting end from the link for the reason that a “client
node” can frequently be movable and uses its link to the base station for those with
its statements through other nodes.
Multiple-­input multiple-­output (MIMO) is a technique for multiplying the
capacity of a radio link using many transmitting and receiving antennas to utilize
multipath propagation. The advantages of using MIMO are increasing link capacity
and spectral efficiency. Here we are dealing with multiple radio channels, i.e., the
WiMAX-­based WSN system will transmit and receive the data through multiple
radio channels. Multiple transmitters can transmit data for a particular receiver at a
time; such a technique is called multiuser MIMO [1]. By using multiple transmit-
ting and receiving antennas or radio channels, MIMO offers extra special degrees of
freedom for data transmission. In a single-­user wireless communication system, it
has been observed that using the MIMO technique can lead to impressive improve-
ment in capacity and link reliability. The MIMO technique has a great potential to
improve the throughput, delay, and jitter performance [2]. MIMO plays a significant
role in any wireless communication system. WiMAX uses the MIMO technique in
terms of multiple radio channels, which results in an improvement in the perfor-
mance of WiMAX-­based systems.
In today’s era of wireless communication systems, the strength and the quality
of the received signal by the user is depending upon various factors. Similarly, in
the transceiver system, there are several factors that govern the signal strength and
quality. Some of these factors are modulation technique, data rate, coding scheme,
power constraint, path loss factor, and so on. Now in order to satisfy the need for
effective signal transmission and to gain a good signal quality at the receiver or
user end, the aforementioned listed parameters have to be adjusted according to the
channel parameters. The novel technique employing this strategy is called the link
adaptation technique.
In the link adaptation technique, various constraining factors are adjusted or
adapted according to the required radio link parameters to provide good-­quality sig-
nals. Recently, an alternative link adaptation technique called AMC has come up to
improve the overall system capacity. AMC allows matching of modulation and cod-
ing method along with protocol and signal parameters to those with the conditions
on the radio link such as path loss, sensitivity of the receiver, power margin of the
transmitter, and so on.
The process of AMC is a dynamic one as the protocol and signal conditions on
radio link alter frequently. The main purpose of AMC is to maintain an acceptable
 Design and analysis of wireless sensor network  237

bit error rate as well as to make more effective use of channel capacity. Various
transmission parameters that can be adapted are data rate, coding rate, error prob-
ability and transmitted power.
Other methods can be the assumption of approximately the same channel from
TX to RX and from RX to TX in the time-division duplex method. Using link state
information that is present in the TX adaptive modulation systems provides an
improvement in the rate of transmission.
The AMC concept. As per the AMC concept, here the modulation format is
matched according to the SNI ratio for end customers. Here, the channel is made
approximately constant by selecting small time–frequency bins, and thus for each
time slot, a separate channel is represented by each of these time–frequency bins.
The efficient technique will be allowing only one user having the best channel to
transmit in each of the parallel channels. For lower values of SINR, the modulation
technique used is QPSK with a smaller constellation size of the signal to provide
reliability and robust transmission. As the SINR values increase, the modulation
techniques with a bigger constellation size of the signal are employed like 8-­PSK,
16-­QAM, and 64-­QAM. For higher values of SINR, the reason for using bigger
size of the constellation of the signal is to provide significant modulation rates with
lesser values of error probability.
MIMO promises a significant increment in throughput and ranges of wireless
communication without any increase in transmit power. A MIMO system relies on
techniques such as spatial multiplexing, transmit diversity, and beamforming to
improve the quality of transmission, data rates, and received signal gain as well as
to reduce interference. Assume a communication system with ηт TX antennas and
ηR RX antennas.
‍ r t = Hs t + v t ‍
 T
Here, rt = ‍ r1t r2t ...rnRt ‍is the receive side signal on time moment t,
1 2 nT T
st ‍= s t , s t , ..., s t ‍is the transmitted signal, and vt is AWGN with unit variance
and uncorrelated among the ηR RX antennas. RX antenna i receives a superposi-
tion of every transmitted message as of TX j, weighted by the link reply, and a few
AWGN is added.
Here, ηR‍‍ηT transition matrix is completed from elements hi,j as given:
0 1
h1,1 ... h1,nT
B C
H=B @ ... .... ... C
A
hnR ,1 ... hnR ,nT
‍ ‍
Here, hi,j represents the complex link coefficient among the jth TX antenna and the
ith RX antenna.
The transmitted power is given by

‍ P tj = P
T ‍
or j=1, 2,…………………., ηт
238  Intelligent network design driven by Big Data analytics

Recently, the requirement for multimedia facilities with large superiority of ser-
vice requirements has been increasing [2]. The speed of the network depends on
multipath propagation and path loss. The key difference between WSN and MANET
is the high-­speed models, fast-­changing topology, and also capability of mobility
prediction. The 4 × 4 MIMO system gives superior performance in the presence of
interference. Also, it improves system’s capacity.
The sending and receiving sections of the antenna are used to suppress the
multi-­antenna fading channel. It can take advantage of MIMO channels to improve
the wireless channel and double the ability exclusive of growing the bandwidth and
transmit power of the antenna. To defeat the troubles of the channel impairments, we
used adaptive modulation for MIMO systems. AMC and MIMO increase the infor-
mation rate of the system. The information of link situation information at the TX
is serious for multi-­user MIMO, while it is not important for single-­user MIMO [3].
In any unplanned network, the nodes are not well known about the topology of
the network. Reactive RP is a bandwidth-­efficient on-­demand routing protocol for
ad hoc networks. The examples of reactive RPs are Ad hoc On-­Demand Distance
Vector (AODV), Ad hoc OnDemand Multipath Distance Vector (AOMDV), and
Dynamic Source Routing (DSR).
In case of proactive RP, each node has either one or more than one table that
shows the whole network topology. These tables are regularly updated so as to pro-
vide advanced information about routing from every node to others. Destination
Sequenced Distance Vector (DSDV) is the proactive RP.
The information regarding the topology is transmitted only on-­demand by the
nodes. AODV uses the shortest path or fastest path for transmission of data. AODV
RP is used to find roots from the transmitter to the receiver only on insist, i.e., only
when the source node wants to transmit data packets, only then the root is found.
This is unipath routing protocol. If the root to the required destination is not avail-
able, then the source node sends RREQ to every neighboring node into the system
as a result of which it gets many roots to the different destinations from just one
root request RREQ message. Ten AODV utilizes target string number for the deter-
mination of up-­to-­date lane and for finding a fresh way to the receiver. The source
node receives RREP packets from intermediate nodes that have a valid root to the
destination node or else the root replies packets, i.e., RREPs are directly sent by
the receiving node to the transmitting node. But in between if the path breaks and
the intermediate node identifies it, then it informs the end nodes about the break-
age of the path by sending a Root Error message. As a result, the corresponding
entry is deleted from the table by the end node. Again now the source node initiates
the search of a new path with the new previous destination sequence number and
new broadcast identifier. For maintenance of roots in AODV, periodic exchange
of HELLO messages is done. Every time when RREQ messages are broadcasted
by source nodes to its neighboring nodes, a reverse path is set up, and due to this
reverse path, a unique ID is allotted. Every node will check the address of the initia-
tor and this unique ID and rejects the message if it had processed that request. The
AODV routing protocol creates a problem during the transmission of data in node-­
to-­node communication [4]. But at the same time, AODV effortlessly overcomes the
 Design and analysis of wireless sensor network  239

counting to infinity and Bellman-­Ford problems, and it also provides rapid conver-
gence whenever the ad hoc network topology is changed [5].
AOMDV RP is used for sensor networks also. This is basically used for link-
ing disjoint paths. As the nodes receive a duplicate root advertisement message, it
denotes another route to the receiver.
AOMDV is used for finding node disjoint roots as well as link disjoint roots.
For finding node disjoint roots, every RREQ coming through another nearest node
of sources presents a node displace lane. For finding several connections to displace
roots, the receiver gives a reply to copy of RREQs. As the primary jump is finished,
the RREPs go through opposite routes and they are node displace and therefore
link put out of place. The paths of every RREP might go through the same midway
node at some point except every one of them gets a dissimilar opposite route to the
transmitter node in order to make sure disjointness. AOMDV has been verified to be
a superior protocol that uses multipath routes. Advantages of AOMDV RP: It is the
distributed protocol to discover link disjoint paths. It reduces overhead by providing
multiple paths.
Disadvantages of AOMDV routing protocol:
It has additional overhead for route discovery for RREP. Because of periodic
route discovery it consumes extra bandwidth.
It emphasizes the issue of improving road safety and transport effectiveness
through the use of WSN. Here the authors have considered the issue of safety in
vehicular communication and for safety-­related applications in WSN; they have sur-
veyed the recent approaches and protocols in WSN. The authentic time route devel-
opment algorithm is used to ease transfer congestion in city areas. First, the authors
have established a hybrid intelligent transportation system and then proposed a novel
lane preparation algorithm that outperforms the conventional scattered lane prepa-
ration algorithms for spatial utilization. Furthermore, Miao Wang et al. designed
an efficient coordinated charging strategy. Through this strategy, the authors have
achieved improvement in the energy utilization and reduction in the electric vehi-
cles’ journey charge: the availability of IPTV services over WSN, communication
scenario to verify wireless communication performance and its operating reliability
record duplication as a scheme for data division in WSN and also compared various
strategies of database replication, and the location Verification Protocol for NLOS
(non-­line-­of-­sight) conditions in WSN. Through simulation results, the authors have
proved that the NLOS condition can be overcome by using the location verifica-
tion protocol among the neighboring vehicles, and thus the integrity of localization
services for WSN can be secured: the analyzed WSN connectivity in case of limited
RSUs (road-­side units) deployment as well as provided for enhancement also.
As the path request packet will propagate in the network, if the destination gen-
erates the route reply, then it will put the path documentation into the path respond
packet from the route request packet. Figures 11.1–11.5 show the path response by
the receiving node itself. Adding new metrics and making few changes in the opera-
tion of DSR protocol using the fuzzy interface system increase the performance in
real-­time applications. In some of the applications, the DSR protocol lags behind
compared with other reactive routing protocols because if any source has more than
240  Intelligent network design driven by Big Data analytics

Figure 11.1   Application of WSN

one route in its cache, which route to choose will totally depend on the source.
When energy efficiency is considered, the DSR protocol lags behind compared with
other reactive routing protocols. The reason behind that is node mobility and node
failures.
Advantages of DSR RP:
Paths remain only between nodes that require transmission.
Disadvantages of DSR RP:

Figure 11.2   ormhole attack

 Design and analysis of wireless sensor network  241

Figure 11.3   Without wormhole attack animator situation

Packet header extents with path length due to source routing.

The middle node may transmit path respond using a stale cached path, thus pol-
luting additional caches.
Here, they have focused on highway WSN scenarios and show that even with
a small number of RSUs along the highways, the performance is significantly
improved. The authors have put forward a survey of MAC (Media Access Control)
protocols for V2V (vehicle-­to-­vehicle) communications in WSN. The authors have
also focused on the benefits and limitations of the various MAC protocols proposed
in the survey. In addition to this, they have highlighted some of the challenges of
the past proposed MAC protocols and the need for novel solutions. They explained
the implementation of the new 17 routing protocol for the WSN environment with
handover mechanism and got good results in conditions of speed than other routing
protocols. It is used in performance evaluation for multicast transmission in WSN.
The new communication strategy is based on conditional transmission for highly
dynamic networks like WSNs. The basic idea used here is sending a message with

Figure 11.4   Network animator situation with designated nodes

242  Intelligent network design driven by Big Data analytics

Figure 11.5   Throughput of the simulation system

some condition used for either reception or retransmission instead of transporting

addresses or positions. The authors have also compared the performances of proac-
tive and reactive protocols. The authors thoroughly elaborated WSNs and discussed
its challenges and future perspectives. They have discussed various applications
of WSN as well as suggested measures for further improvement. Furthermore, the
authors have determined MTR (Minimum Transmission Range) of various road
configurations in WSNs to provide better connectivity.
DSDV has loop-­free paths, requires less convergence time, and is more dynamic.
At every node, topology information is maintained with the help of tables. DSDV
routing protocol can be used for VANET as well as MANET applications. This
protocol is simpler and easy to implement than the protocol using vehicle position
information, which requires the installations of related devices onto all vehicles on
the road. DSDV finds routes in progress for every transmitter and receiver pair and
time-­to-­time exchange topology information to sustain them.
Entry of increasing sequence digit tags helps in keeping the routing table
updated. As a result of which, problems of loop formation, counting to infinity hur-
dle, are avoided and the convergence time is reduced. The routing table is updated
by the destination node also by entering a fresh sequence digit that is superior to the
prior sequence number.
There are two types of methods for updating the routing tables at each node
in DSDV, which are incremental routing table update and full-­dump routing table
update. In case of incremental routing table update, nodes do not notice a consider-
able change in local network topology, and this update process takes a single net-
work data packet unit. In case of full-­dump routing table update, nodes notice a
considerable change in local network topology and it may take more than one net-
work data packet unit.
Advantages of DSDV RP:

• Loop-­free paths are guaranteed by the DSDV protocol.

• DSDV has reduced the count to infinity hurdle to a great extent.
 Design and analysis of wireless sensor network  243

• Incremental updates can be used instead of full dump updates to avoid extra
traffic.

Disadvantages of DSDV RP:

• Multipath routing is not supported by DSDV.

• It is hard to keep the routing table information for a huge network, as this would
lead to overhead resulting in consumption of more Bandwidth.

They designed a high-­speed address generator scheme required for address gen-
eration in deinterleaver of the WiMAX receiver system. Here, they have proposed
a novel Application-­Specific Integrated Circuit-­based design for address generator,
and its modeling has been done using VHDL. The authors have worked on Wi-­Fi
and also WiMAX technologies. Here the authors have evaluated a multi-­vehicle to
infrastructure WSN by using Wi-­Fi for V2V communication and using WiMAX
for the vehicle-­to-­infrastructure communication. They analyze the WiMAX perfor-
mance of efficient wireless channels using image and speech transmission. Here
they have thoroughly explained the concept of WiMAX system modeling with a
proper selection of wireless channels like AWGN, Rayleigh, Racian, and so on so
as to control the BER. The operational inference on WSN is 802.16e (WiMAX) and
802.11 p.
Then the authors have analyzed the performance of Wireless Broadband. Here
they observed that a large portion of the delay in the handover process is due to the
deep computing process occurring during authentication process. WiMAX topology
shows that those data rates humiliate with the increase in the distance greater than
10,000 m. The authors proved that the independent MCS level is better than others:
the performance of proactive RP in WSN over TCP and CBR connections. The
performance is analyzed from parameters like PDR, PLR, and so on. Similar work
of performance evaluation of WSN routing protocol is done by Nicholas et al. in the
field of large-­scale urban environment. The routing protocol used for performance
evaluation is GPSR, Vehicle-­Assisted Data Delivery, and LOUVRE. WiMAX
allows more number of users in a short coverage area. While barriers are there, the
genuine speed may be below 20 Mbps, but WiMAX can offer safe relief of data and
hold mobile subscribers at vehicular mobility. IEEE 802.16e is designed to both
achieve high-­speed data services and allow mobile users with broadband wireless
access solutions. WiMAX enables higher mobility for high-­speed data applications.
The number of IFFT and FFT vectors decides the number of subcarriers gen-
erated for the given OFDM system. For an OFDM symbol, the orthogonality of
subcarriers is given by
fk = k
k = 0, 1, ......, N  1 (11.1)
‍ TMC ‍
Here, T 1 is the intercarrier spacing
‍ MC ‍
k’ is the number of subcarriers whose frequency is to be calculated.
The corresponding kth subcarrier at frequency ‍fk ‍can therefore be written as:
244  Intelligent network design driven by Big Data analytics

‍ Ik (t) =ej2fk t ‍ (11.2)

For N subcarriers of OFDM symbol after being modulated by transmitter:
  N1
P kn
S n = ak ej2 N , 0  n  N  1 (11.3)
‍ k=0 ‍
In 3.3, ak represents data symbols that are properly mapped and are complex in
nature.
In an ideal channel where the transmitter and the receiver of the OFDM system
are properly synchronized, the received subcarriers sequence R(n) is the same as the
transmitted subcarriers sequence, i.e., R(n)= S(n).
So, under these ideal conditions, the demodulated data for the kth subcarrier are
given by
0
kn
^ P   j2
1 N1 (11.4)
ak0 = R n  e N
‍ N n=0 ‍
0
kn kn
P N1
1 N1 P j2 j2 (11.5)
= ak e N e N
‍ N n=0 k=0
0  ‍ 
0 1
n kk
P B
N1
B 1 P
N1 j2 C
C
= ak B e N C
k=0 @ N n=0 A
(11.6)
P
N1  0

= ak ı k  k
k=0
‍ = ak ‍
The ideal case of the OFDM system in practice does not exist. So, some parameters
need to be considered while designing the OFDM system like channel disturbances,
oscillator effects, and so on. Also due to the delay in the channel, the OFDM sym-
bols may overlap with each other causing inter-­symbol interference (ISI), which
results in a change in 3.4. In OFDM symbols, in order to reduce the ISI in dispersive
channels, guard intervals are inserted. After modulation d-­mapping, the original data
will be reconstructed. In this way, the data transmission and reception process is car-
ried out in WiMAX. The technical overview of the WiMAX is given in the follow-
ing. WSN is the application of MANET that improves traveler’s comfort. WiMAX
also provides high-­level security for data transmission and reception in MANET as
well as WSN.
Each vehicular node is capable of receiving, processing, and forwarding the
packet. WiMAX base station has sufficient accessible bandwidth, so at a time, it
can serve a large number of subscribers and also cover large coverage area range.
WiMAX supports the MIMO technique to transmit the data via multiple radio chan-
nels. MIMO transmits multiple signals at a time simultaneously. WiMAX has more
security than Wi-­Fi. Also WiMAX provides a better throughput for high-­mobility
 Design and analysis of wireless sensor network  245

vehicles. The performance of the WiMAX-­based WSN system can be evaluated by

various techniques and protocols in which the show of WSN is generally based on
RP’s.
In today’s era of wireless communication systems, the strength and the quality
of the received signal by the user are depending upon various factors. Similarly, in
the transceiver system, there are several factors that govern the signal strength and
quality. Some of these factors are modulation technique, data rate, coding scheme,
power constraint, path loss factor, and so on. Now in order to satisfy the need for
effective signal transmission and to gain a good signal quality at the receiver or
user end, the above-­listed parameters have to be adjusted according to the channel
parameters. The novel technique employing this strategy is called the link adapta-
tion technique.
In the link adaptation technique, various constraining factors are adjusted or
adapted according to the required radio link parameters to provide good-­quality sig-
nals. Recently, an alternative link adaptation technique called AMC has come up to
improve the overall system capacity. AMC allows matching of modulation and cod-
ing methods along with protocol and signal parameters to those with the conditions
on the radio link such as path loss, sensitivity of the receiver, power margin of the
transmitter, and so on.
The process of AMC is a dynamic one as the protocol and signal conditions on
the radio link alter frequently. The main purpose of AMC is to maintain an accept-
able bit error rate as well as to make more effective use of channel capacity. Various
transmission parameters that can be adapted are data rate, coding rate, error prob-
ability, and transmitted power.
Other methods can be the assumption of approximately the same channel from
TX to RX and from RX to TX in the time-division duplex method. Using link state
information that is present at the TX adaptive modulation systems provides an
improvement in the rate of transmission.
A benefit of a MIMO system is that without using extra power, multiplexing
gain can be enhanced. It uses the highest number of retransmissions suitable for
a packet if the earlier transmissions are not successful. Increasing the retry limit
improves the system reliability, at the cost of higher packet delay and lower chan-
nel throughput. It adapts to the advance antenna technology supporting the MIMO
and HARQ to enhance reliability. MIMO can also be used for underground tunnel
networks. The MIMO capacity depends on transmission power and traffic load. The
MIMO technique can be used to send the data for different sports activities at a time.
In mobile communication, the MIMO technique allows frequency use again concept
within every cell.

11.2 Wireless sensor network

Figure 11.6 displays the basic sensor network for exchange in a position of discrete,
light mass nodes, prepared in large quantity to display the condition or gadget by
way of the scale of the fleshly trouble including contamination, heaviness, stickiness,
246  Intelligent network design driven by Big Data analytics

Figure 11.6   Basic stature of WSN

and so forth. The Wi-­Fi sensor system is a collection of SNs. The evaluation of stor-
age, mining, and processing of the subsystem of the Internet SNs consists of sensor
subsystem, dispensation machine, and message scheme.

11.3 WSN application

It has various applications in this field of WSN. The few applications can be the
following.
Military application: WSN is probably an important piece of martial command,
control, communication for calculating intellect, battleground surveillance, investi-
gation, and directing systems.
Locale monitoring: In this locale monitoring, the feeler nodes are organized
in a section where some experience is to be monitored. At what time the feeler
notices the occasion creature observed high temperature and force, and the occasion
is stated to one is a BS.
Transference: Genuine-­time transfer info is existence composed by WSNs to
future supply transport model and attentive drivers of mobbing and transfer problem.
Healthiness application: In the same health application are at the bottom of
intrusions for disabled, including patient monitor, indicative, and medicine direction
in hospices, tale-­monitoring of humanoid physiological information, and tracing and
monitoring medics or patients within a hospice.
Ecological sense: The team of ecological feeler networks has residential to cover
up lots of applications including air pollution checking, forest fire discovery, green-
house checking, and landslide detection. This is the type of application of ecology.
Structural monitoring: Wireless sensors are able to be used to observe the asso-
ciation inside constructions and infrastructures such as bridge flyovers, embank-
ment, and tunnel.
 Design and analysis of wireless sensor network  247

Industrialized monitoring: Wireless feeler networks contain industrialized sup-

port of technology condition-­based repairs as they present important price invest-
ments and allow novel function a laity.
Agricultural sector: This wireless network releases the grower as of the con-
servation of electric wiring in a problematic atmosphere. Irrigation mechanization
enables extra effectual water use and compact desecrate.

11.4 Limitations of WSN
1. They have a very slight loading capacity – a few hundred kBs.
2. They have hesitant processing power – 8 MHz.
3. They work in a small variety – devours a lot of power.
4. They need small energy – obliges protocols.

11.5 Literature survey

In this chapter, the wireless networks are vulnerable to many outbreaks, with an
outbreak recognized as the loss outbreak. The wormhole outbreak is actual prevail-
ing, and avoiding the attack has been established to be very problematic. In such
outbreaks, double or extra malevolent plotting nodes make an advanced level simu-
lated tunnel in the net, which is working to transport packages. It presents an original
belief-­based structure for classifying and unraveling nodes that generate a wormhole
in the system deprived of attractive any cryptographic income. We establish that our
arrangement purposes efficiently in the occurrence of malevolent conspiring nodes
and do not execute any needless situations upon the system founding and process
stage [1].
In this chapter, present the Ad-­hoc network vehicular are predicted in to the out-
come Wi-­Fi knowledge containing of short range, which is grouping of Wi-­Fi. Other
applicants of prolonged reserves wireless machines are cell, and WI-­max. in meant
device offers a pair of radio channel in among trans receiver aimed at the printed and
response of the information thru using the clue of MIMO expertise. Furthermore,
AMC offers a range of variation methods trusting on the sign to finish relation of the
channel. These two schemes offer the massive alteration inside the excellence of the
current network [2, 3] on this section of the paper Zigbee collection tree be dressed
documented Zigbee topology particularly suitable for WSNs impossible to resist
low power and maintenance reduce change since it supports sturdy factor rescue
schedules [4–8].

11.6 Related work

Figure 11.2 shows that the wormhole outbreak includes the only extra nodes and net-
work between them. The outback node-­sized applications or statistics one recover
248  Intelligent network design driven by Big Data analytics

Table 11.1   Simulation constraints

Constraint Number
Frequency 2.4 GHz
Bandwidth 20 MHz
Transmission model TRG propagation
Mac WiMAX
Nodes 10
Period 10 s
Zone size 500 m × 500 m

and amount them to an additional remotely placed node that allots them neighbor-
ing [9–11]. They can then transmit out several types of bouts in contradiction of
information circulating flows including discriminating decreasing. The strong factor
also displays themselves or masks in a manner. The preceding is showing or naked
wormhole outbreak, on the equal although as the concluding is a covered or close
to one [12–14].

11.7 Methodology

Simulation parameters are important for the design of any system that provides the
information about the nature of the wireless system. Table 11.1 shows the various
simulation constraints used to design the wireless communication system. They are
decided on the basis of applications that are targeted [15–33].
The QoS parameters of vehicular ad hoc networks are important to evaluate
the WSN system. Different QoS parameters used in this WSN system are explained
below.

11.7.1 Throughput
Throughput is the average rate of successful packet transfer between the transmit-
ter and receiver of any system over a communication channel. Throughput of any
network should be as high as possible. Its unit is bps.
Throughput = (Total packets received * Size of packet * 8)/(Time taken for
transmission of all packets).

11.7.2 Delay
Delay is the time taken by the packets to transfer from TX to RX. Delay affects due
to channel conditions, traffic, improper routing, and so on. Delay is an important
performance parameter of any network. Delay shows how fast your data transmis-
sion process is carried out. So delay should be as low as possible. Delay is measured
in seconds. It is calculated by the following expression.
Delay = Packet receive time – Packet send time
 Design and analysis of wireless sensor network  249

Table 11.2   Result of middling value

Parameters Quantity
Throughput 1575.204 bps
PDR 77.76%
Delay 1.69×10ˉ5 s

11.7.3 Packet delivery ratio

It is defined as the percentage of packets successfully received to RX. To improve
PDR, the system should have a low computational requirement and an alternate
route should be available if the current link has broken. The packet delivery ratio
shows the percentage of packets received at the receiver. So, the packet delivery
ratio should be maximum.
Packet Delivery Ratio = ((Total packet received/ Total packet sent) * 100)

11.7.4 Design of WiMAX-based WSN system

This WiMAX-­based WSN system is designed using NS 2.34. The designing of the
WiMAX-­based WSN system is explained in the following.

• The response of any ad hoc network is depending on its simulation parameters.

To design the WiMAX-­based WSN system, the WiMAX parameters play an
important role in data transfer. The frequencies used for WiMAX-­based WSN
systems are 2.5 and 4.5 GHz with a bandwidth of 20 MHz. The WSN system
consists of a number of vehicular nodes depending on the application.
• Create a new simulator with trace and NAM file in the write node. NS 2.34 exe-
cutes the different events on the defined time. So, each event should save in the
trace and NAM file, so that both files must be in write mode. Trace file consists
of the information about packet send, receive, drop, and forward in the network.
It also consists of event time, coordinates of the vehicular nodes, energy, packet
size, routing protocol, and IP address of the TX and RX vehicular node. NAM
file is the Network AniMator that provides the network animation with all the
events.
• Here the new trace format of the wireless communication system is used because
this provides more information about the different events and vehicular nodes.
• The vehicular ad hoc network changes its topography according to the change
in the position of vehicular nodes so that its topography changes continuously.
So, initialize the change of topology area as a simulation area.
• Initialize cluster number for each domain. Also create channels and base station.
• Here base station is positioned at the middle of the network, so it can control
every vehicular node within its coverage range. To identify the base station in
the network, name it BS.
250  Intelligent network design driven by Big Data analytics

Table 11.3   Comparison of Wi-­Fi and WiMAX

Standard IEEE 802.11 a IEEE 802.11b IEEE 802.11 g IEEE 802.16

Frequency (GHz) 5 2.4 2.4 2.66
Speed (Mbps) 54 11 54 80
Range 50 m 100 m 100 m 50 km
Radio technology OFDM DSSS OFDM (64 OFDM (256
channels) channels)
Primary application Wireless LAN Wireless LAN Wireless LAN Broadband
Wireless
Access

• Create vehicular nodes according to the real-­ time vehicular system. Each
vehicular node has its own source and destination locations. Both the locations
should be within its simulation area. The vehicular nodes are moving at some
speed, so assign different speeds to each vehicular node. Give different labels
to each of the vehicular nodes. Here the vehicular nodes are labeled as V1, V2,
V3, and so on.
• Create source traffic. In the WSN system, communication takes place in three
different ways, i.e., V2V, base station to vehicle, and vehicle to base station.
Any vehicular node can communicate with other according to its requirement.
User Datagram Protocol (UDP) agent is used for transmission of data from
transmitting node. Connect UDP agents to all transmitting nodes.
• Constant Bit Rate (CBR) agent traffic source is attached to UDP with a packet
size of 1,000 bytes and different packet interval times to each of the source
vehicular nodes. UDP1 and CBR1 are attached to the same vehicular node so
that it will be transmitter or source node.
• Create the NULL agent to sink traffic. Some of the vehicular nodes are receiv-
ing nodes, so connect NULL agents to all receivers. The NULL agent will ter-
minate the data to that respective vehicular node.
• Attach two vehicular nodes in which agents UDP and NULL are attached to
make pair of transmitter and receiver so that the transmitter node can transmit
the data to the receiving node.
• Start and stop the CBR traffic within the simulation time.

11.8 Related results

The function of NAM is to show the animation of static and dynamic nodes with
packets transfer, packets loss, position of the nodes, and simulation time scale. NAM
also consists of forward, reverse, and stop functions that can be used to see any event
that occurred during the simulation at any time instant. The NAM scenarios for vari-
ous environments are given in the following.
 Design and analysis of wireless sensor network  251

Figure 11.7   Delay among nodes

The situation without wormhole attack animator is shown in

The calculated MANET contains 12 nodes and one base station. The network
continuously transmits data or packets or data over. Communication takings resi-
dence the portable nodes over Base Station Circles displays that the part of coverage
system and the area size is 500 × 500. Figure 11.4 shows the same parameters but
in a different way for the zooming animator scenario; it also presents the 10 nodes
including one base station.
Figure 11.5 displays the speed of the modeled network in which diagram is
planned among simulation time (s) and throughput (bps).
Figure 11.7 displays the delay diagram. The extreme delay is 0.0015 and the
lowest delay is intended as 0. The replication time busy to scheme these standards
are since 0 to 10 s.
Differentiating the overall sum of packets is traditional on the output nodes.
As indicated in Figure 11.8, the package’s transfer margin is 1.69 ms. The ordinary
price of PDR indicates the model time in seconds; this is the 0–12 range and PDR is
measured in ratio, and the all-­out variety of the PDR is 100%.
Figure 11.9 shows the middling value of outcomes. The normal throughput
range is 1,575.204 bps, which is to be intended. The PDR is 77.76%, and the delay
is 1.69×10 −5 s, which is the last reading in Table 11.2. The comparison of Wi-­Fi and
WiMax is given in (Table 11.3).

11.9 Conclusion

The designed WSN system is useful for transportation systems as well as indus-
try automation also. This system provides better performance than the exist-
ing system. The various simulation parameters show that the designed system
provides a high speed of the network for transportation systems as well as for
other applications.
252  Intelligent network design driven by Big Data analytics

Figure 11.8   PDR network

11.10 Future scope

The investigation carried out in this book chapter leaves an ample scope for the
extension of the WiMAX-­based WSN system for various environment networks.
The performance of the WiMAX-­based WSN system can be improved in the future
by using the following ways.
The use of hybrid routing protocols in the WiMAX-­based WSN system can
increase the performance of the network.
Various power reduction techniques can reduce power utilization at the base
station.
By using various low-­energy-­consumption algorithms, the utilization of energy
per vehicular node can be reduced.
One can design a WiMAX-­based WSN system for a large coverage area net-
work by using multiple relay stations to improve the efficiency.

Figure 11.9   Relation of modulation techniques and throughput

 Design and analysis of wireless sensor network  253

It can be concluded that the work under investigation on the design of the
WiMAX-­based WSN system can be extended toward various applications in the
area of intelligent transportation systems. The practical development of an improved
WiMAX-­based WSN system is one of the major potential research directions in the
future.

References

[1] Parmar A., V.B. V. ‘Detection and prevention of wormhole attack in WSN
using AOMDV protocol’. 7th International Conference on Communication,
Computing and Virtualization; Maharashtra, India, 2016. pp. 700–07.
[2] Dorge P.D., Dorle S.S. ‘Design of WSN for improvement of QoS with differ-
ent mobility patterns’. 6th International Conference on Emerging Trends in
Engineering and Technology; Nagpur, India, 16-­18 Dec; 2013.
[3] Dorge P.D., Dorle S.S., Chakole M.B., Research Scholar, G. H. Raisoni
College of Engineering, Nagpur, India 'Implementation of MIMO and AMC
techniques in wimax network based vanet system'. International Journal of
Information Technology and Computer Science. 2016, vol. 8(2), pp. 60–68.
Available from http://www.mecs-press.org/ijitcs/v8n2.html
[4] Shende S.F., Deshmukh R.P., Dorge P.D. ‘Performance improvement in
ZigBee cluster tree network’. International Conference on Communication
and Signal Processing (ICCSP); Chennai, India, 6-­8 Apr; 2017.
[5] Pochhi R.D., Deshmukh R.P., Dorge P.D. ‘An efficient multipath RP for cog-
nitive AD hoc networks’. International Journal of Advanced Electrical and
Electronics Engineering. 2012, vol. 1(3), pp. 1–7.
[6] Meshram S.L., Dorge P.D. ‘Design and performance analysis of mobile Ad
hoc network with reactive RPs’. International Conference on Communication
and Signal Processing (ICCSP); IEEE, Chennai, India, 6-­8 Apr; 2017.
[7] Pandilakshmi S., Amar R. ‘Detecting and prevent the Wormhole attack us-
ing customized evolution’. International Journal of Innovative Research &
Studies. 2018, vol. 8(4), pp. 1–7.
[8] Ghormare S.N., Sorte S., Dorle S.S. ‘Detection and prevention of Wormhole
attack in WiMAX based mobile Adhoc network’. Second International
Conference on Electronics, Communication and Aerospace Technology
(ICECA); Coimbatore, India, 29-­31 Mar; 2018.
[9] Siva Ram M.C., Manoj B.S. Ad hoc wireless networks architecture and pro-
tocols, upper saddle river, NJ: Prentice Hall PTR; 2004.
[10] Gupta S., Kar S., Dharmaraja S. 'WHOP: wormhole attack detection pro-
tocol using hound packet'. Presented at The International Conference on
INNOVATIONS Technology; Abu Dhabi, United Arab Emirates. IEEE,
[11] Hu Y.-C., Perrig A., Johnson D.B. ‘Wormhole Attacks in Wireless Networks’.
IEEE Journal on Selected Areas in Communications. 2016, vol. 24(2), pp.
370–80.
254  Intelligent network design driven by Big Data analytics

[12] ChiuH.S., Lui K.S. ‘DelPHI: wormhole detection mechanism for ad hoc
wireless networks’. 1st International Symposium on Wireless Pervasive
Computing; Phuket, Thailand, IEEE, 2006. pp. 6–11.
[13] Chaurasia U.K., Singh V. 'MAODV: modified wormhole detection AODV
protocol'. IEEE. 2013, pp. 239–43.
[14] Dorge P.D., Dorle S.S., Chakole M.B., Thote D.K. ‘Improvement of qos in
WSN with different mobility patterns’. International Conference on Radar,
Communication and Computing; Tiruvannamalai, India, IEEE, 2012. pp.
206–09.
[15] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: eenhanced
multi-­hop energy efficient distributed protocol for heterogeneous wire-
less sensor network’. Fifth International Conference on Communication
Systems and Network Technologies; Gwalior, India, IEEE, 2015. pp.
194–200.
[16] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization technique for
distributed localized wireless sensor network’. International Conference
on Issues and Challenges in Intelligent Computing Techniques (ICICT);
Ghaziabad, India, IEEE, 2014.
[17] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India, IEEE, 2021. pp. 1–6.
[18] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems; Uttarakhand, India, IEEE, 2020. pp. 327–36.
[19] Kumar S., Trivedi M.C., Ranjan P., Punhani A., et al. Evolution of Software-­
Defined Networking Foundations for IoT and 5G Mobile Networks. Hershey,
PA: IGI Publisher; 2020. p. 350.
[20] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[21] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[22] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
Personal Communications. 2021, vol. 10(3).
[23] Kumar S., Cengiz K., Trivedi C.M., et al. ‘DEMO enterprise ontology with
a stochastic approach based on partially observable Markov model for data
aggregation and communication in intelligent sensor networks’. Wireless
Personal Communication. 2022.
 Design and analysis of wireless sensor network  255

[24] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering. 2019, vol. 8(6), pp. 2278–3075.
[25] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based Sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[26] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring
using machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer2022.
[27] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[28] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium, Publisher: The Electromagnetics
Academy; Prague, Czech Republic, 2015. pp. 2363–67.
[29] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
International Conference on Computational Intelligence and Communication
Networks (CICN); Jabalpur, India, IEEE, 2016. pp. 79–84.
[30] Sharma A., Awasthi Y., Kumar S. ‘The role of Blockchain, AI and IoT for
smart road traffic management system’. IEEE India Council International
Subsections Conference (INDISCON); Visakhapatnam, India, 3-­4 Oct; 2020.
pp. 289–96.
[31] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. 10th International
Conference on Cloud Computing, Data Science & Engineering (Confluence);
Noida, India, IEEE, 2020. pp. 63–76.
[32] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation of
fault tolerance technique for internet of things (iot)’. 2020 12th International
Conference on Computational Intelligence and Communication Networks
(CICN); Bhimtal, India, IEEE, 2020. pp. 154–59.
[33] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 4th International Conference on Information Systems and
Computer Networks, ISCON 2019; Mathura, India, IEEE, 2019. pp. 250–55.
This page intentionally left blank
 Chapter 12
A review of edge computing in healthcare
Internet of things: theories, practices
and challenges
Shamik Tiwari 1 and Vadim Bolshev 2,3

The pandemic has forced industries to move immediately their critical workload to
the cloud in order to ensure continuous functioning. As cloud computing expansions
pace and organisations strive for methods to increase their network, agility and stor-
age, edge computing has shown to be the best alternative. The healthcare business
has a long history of collaborating with cutting-­edge information technology, and
the Internet of Things (IoT) is no exception. Researchers are still looking for sub-
stantial methods to collect, view, process, and analyse data that can signify a quanti-
tative revolution in healthcare as devices become more convenient, and smaller data
becomes larger. To provide real-­time analytics, healthcare organisations frequently
deploy cloud technology as the storage layer between system and insight. Edge com-
puting, also known as fog computing, allows computers to perform important analy-
ses without having to go through the time-­consuming cloud storage process. For this
form of processing, speed is key, and it may be crucial in constructing a healthcare
IoT that is useful for patient interaction, inpatient treatment, population health man-
agement and remote monitoring. We present a thorough overview to highlight the
most recent trends in fog computing activities related to the IoT in healthcare. Other
perspectives on the edge computing domain are also offered, such as styles of appli-
cation support, techniques and resources. Finally, necessity of edge computing in the
era of Covid-­19 pandemic is addressed.

1
School of Computer Science, University of Petroleum and Energy Studies, Dehradun, India
2
Laboratory of Power Supply and Heat Supply, Federal Scientific Agro engineering Center VIM,
Moscow, Russia
3
Laboratory of Intelligent Agricultural Machines and Complexes, Don State Technical University,
Rostov-­on-­Don, Russia
258  Intelligent network design driven by Big Data analytics

12.1 Introduction

The IoT has transformed how healthcare solutions operate, and the industry has seen a
significant shift away from on-­premise hardware and software and towards cloud com-
puting [1]. IoT has spread across a variety of markets, catering to customers on a global
scale. From smart voice assistants to smart homes, brands are diversifying their offerings
and experimenting with new designs to increase consumer engagement.
IoT devices include gadgets, sensors, actuators, appliances and machines, which
are designed for particular purposes and may broadcast data over the Internet or other
networks. The IoT is introducing a new layer of complexity to the task of analysing an
ever-­growing mountain of data in order to improve healthcare [2]. Those who are not
physically available in a health institution can use IoT devices to accumulate health
indicators such as heart rate, temperature, blood pressure, glucose level and more, reduc-
ing the necessity for patients to travel to clinicians by collecting health data themselves.
While much of the conversation around big data has focused on the possible shortcom-
ings of Electronic Health Records (EHRs) and the major challenges of intuitive, efficient
decision support, the IoT is adding another layer of complexity. Traditional information
governance solutions are inadequate to manage this jumbled, unstandardised and poorly
defined amount of data, and they are seen undesirably by overloaded, dispassionate
health practitioners [3, 4]. Furthermore, the massive size of Patient-­Generated Health
Data (PGHD) produced each day presents huge problems for already overburdened ana-
lytics infrastructures that lack the ability to handle the avalanche of big data coming their
way. If information from his wearable sensors don’t reach the treatment station in time, a
patient in the acute care unit has only minutes before a dip in vital signs leads into a dev-
astating crash. Healthcare organisations frequently utilise cloud-­based solutions as the
process, service or storage layer between system and insight to allow what is now known
as ‘real-­time analytics’. Data is bulk-­uploaded to the cloud, and associated components
are discovered and dragged back down into a server for analytics before being submitted
to a user interface visualisation [5]. In a perfect world, the operation will only take a few
minutes but patient life and quality-­of-­care judgements cannot be compromised due to
time lag. The solution could be found in the edge or fog but not in the cloud. This work
provides the details of fog computing including its applications to healthcare IoT. The
rest of the chapter consists of sections on cloud computing in healthcare and its limita-
tions, fog computing and its advantages over cloud computing, role of IoT in healthcare,
practice of fog computing in healthcare, significance of machine learning in healthcare,
integrated impact of IoT, machine learning and fog computing in healthcare, modeling
and simulation tools for fog computing, necessities of edge computing in pandemic era
and conclusion in that order.

12.2 Cloud computing in healthcare and its limitations

Cloud computing is a computational model that uses the Internet to connect comput-
ers, data centres, processors, servers, wired and wireless networks, storage, develop-
ment tools and even healthcare applications. A cloud service provider may take care
 A review of edge computing in healthcare internet of things  259

Figure 12.1   Some scenarios for cloud-­based healthcare system

of any or all of these needs, including providing equipment, training and ongoing
maintenance. Cloud computing can be deployed in different manners, depending on
the resources that an organisation demands. The first thing to consider is the deploy-
ment model – private cloud, public cloud, hybrid cloud and multicloud depending
on the aims of the business use case, each deployment type has strengths and limi-
tations. When deciding on a cloud migration plan, an organisation must weigh all
factors [6, 7]. Some scenarios for cloud-­based healthcare system are presented in
Figure 12.1.

12.2.1 Public cloud
A public cloud is a cloud deployment type in which a vendor owns and operates
computational resources that are shared across several tenants through the Internet.
A public cloud is an open system that enables customers to access storage or software
for freely or on a pay-­per-­use basis over the Internet. A public cloud is a large data
centre that provides all of its users with the same services. The services are avail-
able to anyone and are widely used by consumers. Amazon Elastic Cloud Compute
(EC2), Google App Engine, IBM Blue Cloud and Azure Services Platform are few
examples of common public clouds.
260  Intelligent network design driven by Big Data analytics

12.2.2 Private cloud
A private cloud (or corporate cloud) is a cloud computing infrastructure in which all
equipment/software services are devoted to a single client and only that client has
access to them. Private clouds are limited in size and are used by multiple clients
who access virtualised services and draw resources from a separate pool of physical
computing. With internal hosting and firewalls, the private cloud guarantees that
data is private and secure. It also assures that third-­party vendors do not have access
to functional or critical information. Private clouds include Elastra-­private cloud and
HP data centres.

12.2.3  Hybrid cloud

A hybrid cloud is one in which services are deployed across multiple contexts. Since
nearly no one nowadays relies only on the public cloud, hybrid cloud computing
options are becoming increasingly popular. Although hybrid cloud solutions inte-
grate private and public resources in a variety of ways, they frequently use industry-­
standard technologies like Kubernetes to coordinate container-­based services.
Google Anthos, Azure Stack, AWS Outposts, Azure Arc and VMware Cloud on
AWS are other examples.

12.2.4 Community cloud
The phrase ‘community cloud computing’ refers to a shared cloud computing
service environment geared towards a small number of businesses or employees.
Technically, community cloud is a multitenant platform that is only available to a
limited number of customers. It can be shared by organisations with similar com-
puting concerns and mutual interests. Ventures, corporate organisations, research
groups and tenders are the greatest candidates for this form of cloud computing.
This allows community cloud users to understand and analyse market requirements
upfront.
Many EHRs are actually housed on conventional client–server architectures.
IoT has also aided in the simplification of operations in this area, making the process
much more effective and patient-­centric than it was a decade ago. Healthcare activi-
ties can be made much more convenient and cost-­effective by implementing cloud
computing solutions [8].
The cloud provides on-­demand computing by deploying, accessing and utilis-
ing networked information, software and services using cutting-­edge technology.
However, there are certain drawbacks of cloud computing in healthcare.

• Connectivity: While cloud computing requires connectivity to the cloud, some

IoT systems must function even if the link is temporarily unavailable.
• Bandwidth: Cloud computing assumes that there is enough bandwidth to col-
lect data, but for industrial IoT applications, this may be an excessively opti-
mistic assumption.
 A review of edge computing in healthcare internet of things  261

• Centralised analytics: Cloud computing centralises analytics, thus determin-

ing the system’s lower bound response time. Some IoT applications would be
unable to wait for data to be sent to the cloud, analysed and insights to be
returned.
• Security: Existing data security systems in cloud computing, such as encryp-
tion, have failed to secure data from attackers.

12.3 Edge computing and its advantages over cloud computing

Data processing, analytical and computational skills are brought nearer to the net-
work’s edge with edge computing. These are the so-­called ‘things’ in the context of
an IoT network. There is some ambiguity about the use of these two words in the IoT
industry [9]. Others see a slight difference in that the calculation is performed differ-
ently in a fog or edge scheme, but they are also used interchangeably. Edge computing,
also known as fog computing, enables computers to perform essential analytics without
relying on the inefficient cloud storage method. For this form of processing, speed is
important, and it could be crucial to creating the IoT in healthcare truly effective for
patient interaction, nursing, healthcare management and remote health assistance. Fog
computing is a distributed computing model that sits between cloud data centres, IoT
computers and sensors as an intermediary layer [10, 11]. It offers cloud-­based services
with computing, networking and storage capabilities that can be applied closer to IoT
devices and sensors. Cisco introduced fog computing in 2012 to handle the problems
that IoT applications face in traditional computing environments. IoT devices and sen-
sors, as well as real-­time and latency-­sensitive service specifications, are widely spread
at the network’s edge. Cloud data centres are physically centralised, and they regularly
struggle to meet the storage and handling demands of billions of globally dispersed IoT
devices and sensors. As a result, the network is congested, service delivery is late and
the quality of service is weak [12, 13]. Figure 12.2 presents edge and cloud computing
with end point devices. Table 12.1 assesses cloud computing and edge/fog computing.

12.3.1  Advantages of edge/fog computing

For IoT, big data and real-­time analytics, edge and fogging methods offer numerous
benefits. The advantages of edge computing over cloud computing are as follows:

• Low latency: The latent of edge/fog computing to improve network efficiency

by decreasing latency is its most substantial benefit. The information collected
by IoT edge computing devices does not have to move near as far as it would in
a standard cloud architecture because the data is processed locally or in nearby
edge data centres.
• Low bandwidth requirement: Because data is collected at multiple locations
rather than being sent all at once to a single destination via a single channel,
there are no bandwidth constraints.
262  Intelligent network design driven by Big Data analytics

Figure 12.2   Edge and cloud computing with endpoint devices

• High reliability: Due to a network of interconnected channels, a loss of con-

nection is impossible.
• High security: In a complex distributed system, a large number of nodes,
resulting in excellent security, process data.
• Improved user experience: Users are satisfied because they receive immediate
responses and there are no downtimes.
• Power efficiency: Low-­power protocols like as Bluetooth, Zigbee and Z-­Wave
are used by edge nodes, which make it power efficient.

12.3.2  Disadvantages of edge/fog computing

Although the technology has no apparent flaws, there are a few flaws to be aware of:

• A more complicated approach: Fog is a data processing and storage device

that is applied as a layer.
• Additional costs: Edge devices such as routers, hubs and gateways may be
purchased by companies.
• Scalability: Fog’s scalability is reduced compared with the cloud.
 A review of edge computing in healthcare internet of things  263

Table 12.1    Assessment of cloud computing and edge/fog computing

Parameters Cloud computing Edge/fog computing

Latency High latency in terms of Low latency in terms of
network network
Speed Access speed is high High even more compared
depending on Virtual with cloud computing
machine (VM)connectivity
Processing power Superior and advanced Lesser processing
processing capabilities capabilities
Scalability High Lesser compared with cloud
computing
Storage High Low
Data processing point Central cloud server Located near the sensors on
the devices to which the
sensors are attached, or
on a gateway system
Distance between client and Multiple hops Single hope
server
Security Undefined Possible to define
Vulnerability High chance Low chance
Location alertness No Yes
Data integration Data from many sources can Data from a variety of
be combined sources and devices can
be combined
Geodistribution Centralised Distributed
Responsiveness Low Fair
Support of mobility Limited Available
Type of last mile connectivity Leased line Wireless
Number of server nodes Limited Extensive
Real-­time communications Provided Provided
Purpose It is best for long-­term, in-­ Best suited for real-­time
depth data analysis response analysis that
requires fast analysis

12.4 IoT in healthcare

Due to IoT-­enabled products, remote monitoring in the healthcare business is

now possible, which helps retain patients safe and secure while driving doctors
to deliver superior treatment. As communications with doctors have become
simple, rapid and more successful, patients’ involvement and enjoyment have
improved. Furthermore, remote patient control is beneficial [14, 15]. IoT
devices provide a plethora of new opportunities for healthcare providers to
monitor their patients’ health as well as for individuals to monitor their own
health. As a result, the numerous wearable IoT devices offer a number of bene-
fits and challenges to both healthcare providers and patients. Figure 12.3 shows
264  Intelligent network design driven by Big Data analytics

Figure 12.3    Component of IoT-­enabled healthcare system

components of IOT enabled healthcare system. Some prominent examples of

IoT applications for healthcare are as follows:

• depression and mood monitoring

• glucose monitoring
• ECG monitoring
• PCG monitoring
• remote patient monitoring
• heart rate monitoring
• body hygiene monitoring
• robotic surgery

IoT can play a critical role in smart hospitals, for example. By integrating IoT
technologies into the healthcare sector, it can change almost all of these situations.
Utilising blockchain and smart contracts, inefficient large paper registries can be
substituted by an automated, centralised database. Submissions can be received,
queues can be controlled and staff members may be tracked in real time via smart-
phones with an all-­encompassing processing system. Using blockchain and smart
contracts, any equipment can be continuously controlled and maintained.

12.5 Edge computing in healthcare

The availability of critical data and real-­time data analysis is beneficial to every
firm. These variables, on the other hand, could spell the change between life and
loss in the healthcare. The vast bulk of processing remains to take place in the cloud
or at single-­site data centres. Analysing data from a distance, particularly from the
latter, presents a number of challenges. Congestion of bandwidth, high latency and
low reliability are examples of these issues [16, 17]. Even with today’s 4G LTE net-
works, these problems will still arise as every second counts. Edge/fog computing
seeks to solve these problems by taking data processing closer to the data collection
devices. This is particularly useful in cases where data must be acted on right away,
 A review of edge computing in healthcare internet of things  265

such as in healthcare. There isn’t time to upload it to the cloud and process it when
every second counts.
Improved security, quicker accessibility to real-­time data and transmission effi-
ciency are the three significant benefits of edge computing. Because each edge data
centre handles less data, the security of health records can be improved. Because
the amount of potentially sensitive data in each site is less, malicious hackers will
have a tougher time compromising critical resources or infecting the existing net-
work. Establishing a closed-­loop system in an intensive care unit (ICU) that includes
smart systems to detect acutely ill patient can help healthcare professionals react
in condition more quickly is an example of edge computing in action. In an ICU,
edge computing is performed by attaching sensors to modest, local control systems
that manage computation and transmission. Organisations can realise the following
advantages after they have precisely articulated what they really want to aim from
edge computing and what they have to do to enable it:

• The proximity and efficiency of data

• Integration of IoT data and healthcare provider systems
• Data about patient health that is more accurate and up to date
• Medical gadgets such as insulin pumps, smart glasses and pacemakers are used
to monitor patients
• Telemedicine
• Wearables and application support that monitor a variety of health indicators

12.6 Machine learning in healthcare

Machine learning (ML) is a sort of data analysis that uses artificial intelligence (AI)
to build analytical models. It is a subfield of AI centred on the idea that robots can
learn from data, see patterns and create decisions with minute to not at all human
intervention. ML refers to a system’s potential to ‘learn’ by identifying patterns in
huge datasets. In other words, the ‘solutions’ generated by ML algorithms are statisti-
cal conclusions drawn from very big datasets. The ability to collect, distribute and
deliver data is becoming increasingly important as digitalisation disrupts every sec-
tor, including healthcare [18, 19]. ML, big data and AI will all assist in overcoming
the obstacles that large quantities of data present.
ML is a branch of AI that includes the methods that permit machines to
infer implication from historical data and design intelligent systems. However,
deep learning (DL) is a subset of ML that enables machines to comprehend
extremely complex problems. DL is a relatively new branch of AI based on arti-
ficial neural networks. We can classify DL algorithms as a subsection of ML
since they require data to learn and solve issues. Figure 12.4 provides machine
learning and allied areas.
266  Intelligent network design driven by Big Data analytics

Figure 12.4   ML and allied areas

Healthcare organisations can use ML to meet rising patient demands, optimise

procedures and cut costs. ML advancement at the bedside will assist healthcare prac-
titioners in detecting and treating illness more quickly, with greater precision and
with more personalised treatment [20, 21]. Some prominent applications of ML in
healthcare are as follows:

• Identification of diseases and diagnosis

• Medical imaging
• Drug discovery and manufacturing
• Personalised medicine/treatment
• Predicting diseases
• Medical research
• Smart health records
• Identify relationship between disease and symptoms
• Robotic surgery

In healthcare, AI and ML are still in their infancy. Adoption on a big scale has
yet to occur. In order to be effective in the healthcare industry, AI and ML must have
the support of healthcare medical specialists and doctors.
However, a lot of money is being invested in AI in healthcare, and it is growing
quickly. AI in healthcare is now targeted at improving patient outcomes, balancing
 A review of edge computing in healthcare internet of things  267

Figure 12.5   ML applications in healthcare

the interests of multiple stakeholders, enhancing availability and lowering healthcare

expenditures. Moreover, in the coming years, AI and ML, as well as methods supported
by data science and DL will play a much more holistic role in advancing healthcare.
Figure 12.5 presents machine learning applications in healthcare domain.
DL algorithms are not only remarkably precise but also extremely quick. Studies
have discovered a deep neural network capable of identifying some neurological brain
diseases faster and more accurately than radiologists, such as stroke and brain bleeding.
In few seconds, the system examines the image, evaluates its contents and alarms in
cases of a problematic clinical finding. In neurological settings, ‘time is brain’. In other
words, in the diagnosis of patients’ neurological disorders, a quick reaction is crucial;
any methods that reduce the time it takes to diagnose a patient may result in a bet-
ter outcome. DL is most commonly used in medical imaging. A representative of DL
applications in the medical imaging is provided below. These are only a small number
of applications, and it will continue to grow over time. It gives an indication of the long-­
term influence of DL in the medical imaging sector today.

• Diabetic retinopathy
• Brain tumour
• Alzheimer disease
• Tracking tumour growth
• Breast cancer
268  Intelligent network design driven by Big Data analytics

• Covid-­19-­related lung abnormalities

• Skin cancer, etc.

12.7 Integrated role of IOT, ML and edge computing in

healthcare

Connected healthcare is gaining significant attention from academics, governments,

industry and the healthcare community as smart sensory media, objects, edge analyt-
ics, AI and cloud technologies advance. Patients now need a comprehensive and spe-
cialised smart healthcare system that is customised to their individual health needs,
thanks to the introduction of emerging technology and the quick pace of human life.
Edge/fog computing, in conjunction with 5G networks, AI and cutting-­edge smart
IoT sensors, enables sophisticated, real-­time healthcare applications that are low on
energy and low on latency. While researchers have done substantial development in
the study of edge-­AI and IoT for health services on their own, there has been little
focus on developing cost-­effective and inexpensive smart healthcare services. The
linked healthcare industry’s edge-­AI-­driven IoT has the potential to change many
aspects of our industry [22].

Figure 12.6    I ntegrated role of IOT, ML and edge computing in healthcare. The
edge computing paradigm brings processing closer to physical IoT
equipment, which serves as a vital midway for lowering latency and
conserving bandwidth in the cloud.
 A review of edge computing in healthcare internet of things  269

Following are some case studies where integrated role of IOT, ML and edge/
fog computing in healthcare can be applied. Figure 12.6 discusses integrated role of
IOT, ML and edge computing in healthcare.

12.7.1 Patient care during surgical procedure

Edge/fog computing has the potential to alter operating theatres in hospitals. In the
past, a nurse or two would have to be present in the operating room, maintaining
meticulous records and charting each action being taken by the surgeon and his or
her staff. Software that can log each action is now available, thanks to the deploy-
ment of cameras and other edge computing devices. This helps to save time for
nurses and eliminates the potential for human error. Nurses and surgeons are able to
focus on the patient.

12.7.2 Patient care at home

Most do not have easy access to high-­quality healthcare. From the comfort of their
own homes, edge data centre technologies can enable patients receive better treat-
ments and more accurate diagnoses. While this was formerly assumed to only assist
residents in distant places, the Covid-19 pandemic has educated us that there are a
variety of other reasons that people may be unable to evacuate their families and
seek medical attention in a hospital. Wearable sensors that patients can wear in their
homes benefit from edge/fog computing. These sensors can collect data on patient
health and deliver precise faced by healthcare practitioners as it becomes available.
m-­Health discussions benefit from edge data centre services as well. Drone deliver-
ies of medical supplies and lab specimens are also possible, thanks to cutting-­edge
equipment.

12.7.3  Patient care in ambulance

For in-­hospital physicians and nurses, predicting a patient’s health when they arrive
for care via ambulance is no longer a guessing game. First services used to be able
to relay certain critical information by radio. Today’s cutting-­edge sensors and sys-
tems can send live ambulance surveillance video to doctors, giving emergency room
professionals real-­time patient information. The team can adequately position for
the patient’s presence with this knowledge.

12.8 Modelling and simulation tools for edge/fog computing

As attention in edge/fog computing grows, so does the demand for simulation plat-
forms to facilitate the development and evaluation of edge computing systems. In
many cases, in addition to real-­world solutions, simulations are needed to exam-
ine the behaviour of composite IoT-­edge-­cloud systems or to create modern, effec-
tive data analytics solutions. Several simulators for examining distributed systems,
270  Intelligent network design driven by Big Data analytics

especially IoT and cloud systems are available these days for researchers. Some
prominent tools for simulation of fog and edge computing are listed below [23, 24].

• iFogSim
• FogTorch
• FogTorchPi
• FogNetSim++
• FogBed
• MaxiNet
• EmuFog
• Yet Another Fog Simulator
• IoTSim
• FogExplorer
• RECAP
• EdgeCloudSim
• Sleipnir

iFogSim is the most popular of these tools. iFogSim is a Java-­based open-­source

simulation programme based on Java for simulating fog communications environments.
It was developed at the Cloud Computing and Distributed Systems Lab, University of
Melbourne. It’s features include mobility support and migration management, dynamic
distributed clustering, microservice orchestration and full compatibility with CloudSim
The iFogSim architecture, as shown in Figure  12.7, is made up of layers that are in
charge of different tasks. At the bottom of the layer is the IoT devices, for example,
actuators and sensors that interact with the real world, indicating whether they are data
producers or sinks. To mimic more complex equipments, such as environmental sen-
sors, smart cameras and mobile vehicles, these components can be tweaked to mimic
the effects of collecting data or performing actions. Fog endpoints, according to iFog-
Sim, are any network element that can host application modules. Figure 12.8 presents
iFogSim simulation architecture.

Figure 12.7   Cloud deployment models

 A review of edge computing in healthcare internet of things  271

Figure 12.8   iFogSim simulation architecture (adapted from [25])

12.9 Edge computing in Covid-19 pandemic era

The Covid-­19 pandemic and accompanying lockdowns have prompted significant

adjustments in our lifestyles, as well as exposing major flaws in the broader network
systems that sustain us. More remote work, virtual collaboration and increasing
bandwidth requirements are just a few of the changes [26]. However, the flaws, such
as wired networks that have slowed due to increased load, must be addressed. Due to
the pandemic, organisations have been obliged to hastily migrate critical workloads
to the cloud in order to maintain business continuity. As cloud computing gains pace
and businesses hunt for ways to increase their network, storage and agility, edge
computing has shown to be the best alternative. Every industry is being put to the
test as the globe prepares for the continued effect of the Covid-19 pandemic [27, 28].
As the health industry re-­evaluates their long-­term network needs based on their
experience tackling the Covid-­19 pandemic, edge computing is now emerging as the
forefront and crucial pillar of the network architecture to accommodate this new dis-
tributed workforce and effectively control the rising universe of sensors and devices
at the edge of their networks [29, 30]. Edge computing can support the pandemic era
in following manners.
272  Intelligent network design driven by Big Data analytics

• Creating the self-­contained healthcare centres

• Providing opportunities for new health workers
• Increasing the impact of health Industry 4.0
• Speedy diagnosis of Covid-19 patients
• Distant intensive care of Covid-19 patients in quarantine institutions and home
isolation
• Quick analysis of Covid-19 patient data
• Securing medical records
• Edge/fog computing–based solutions for mobile applications and mobile
payments

12.10 Challenges of edge computing

Data processing for IoT achieves somewhat efficiency due to edge/fog computing.
It adds to the organisation’s coordinated and accurate execution. However, along
with the nice stuff, there are some severe difficulties to deal with. The following are
some of the major issues that arise with implementing edge-­computing technologies
[31–33].

• Data centres acquire higher bandwidth under the conventional asset distribution
model, while endpoints get less. Because edge data processing necessitates a
large amount of bandwidth for optimised and efficient workflow, the dynamics
of edge computing altered intensely. Retaining a balance between the two while
achieving decent performance is the issue.
• Upgrades to edge devices or services might go wrong, resulting in clusters or
devices failing.
• Edge/fog computing raises the significance of the physical setting and comput-
ing environment for data collection and processing. Organisations must have
a position in local data centres to resulting optimal workload and give reliable
result [25, 34–38].
• The set of modules in most servers is distributed and placed high off the ground.
Edge Computing (EC) on the other hand, usually brings all systems closer to
the computational areas. This causes a conflict because the business server must
consider the edge server during computation.
• It becomes more difficult to detect and oversee when firms install ever more
edge nodes to oversee a larger range of processes. Devices may eventually
exceed the edge’s bounds, causing bandwidth saturation and compromising the
security of numerous devices. IoT traffic raises delay as it expands, and when
data is transferred untreated, it might endanger security [39–44].
• Because of the multiple edge receivers located at various distances from the
data centre, troubleshooting and repairing any issues that arise in the framework
necessitate a significant amount of logistical as well as manual input, raising the
cost of maintenance.
 A review of edge computing in healthcare internet of things  273

• Edge devices are frequently ignored by security-­conscious personnel, and many

have inadequate password policies. Hackers have devised advanced methods
for breaching password schemes [45–49].
• Integrated data and network security practices are possible due to a centralised
cloud infrastructure. On the other hand, EC/Fog computing (FC) necessitates
implementing similar policies for remote servers, despite the fact that security
perimeter and traffic conditions are more difficult to evaluate.
• A centralised cloud architecture can be integrated with a system-­wide data loss
protection solution. Edge/fog computing’s decentralised architecture necessi-
tates specialised control and maintenance tools to manage information at the
edge.
• Contain local storage to enable high-­performance and low-­latency data process-
ing, as well as connections to remote storage solutions.
• Data access and storage controls must be handled differently in the edge/fog
computing framework. While centralised infrastructure provides for consis-
tent regulations, edge/fog computing necessitates constant monitoring of each
‘edge’ location.
• While safeguarding patient data may come at a steep price to the sector, it is an
important duty for healthcare to play. It is crucial not just for the protection of
personal patient information but also for ensuring compliance with legislation
[50–52].

In order for ‘the edge’ to become as widespread in the business world as ‘the
cloud’, plenty of technical difficulties must be overcome. Those include production
of small devices with large computing power, software that allows businesses to
remotely control and maintain an unlimited number of edge devices from anywhere
in the globe and additional security techniques and protocols to ensure things pro-
tected. Red Hat, Amazon, Microsoft, IBM, Nutanix and Cloudera, for example, are
all continuously trying to solve these issues and have built their own edge solutions.

12.11 Conclusion

The transition to edge/fog computing solutions is now and will continue to be a

healthcare necessity, particularly as organisations begin to implement post-­Covid
recovery approaches. Living on the edge will soon become the norm in any organisa-
tion, both as a result of and in spite of the pandemic’s impact on healthcare industry.
Existing physical equipment will be transformed from a single-­function activity to
an aggregation of processes, inputs and outputs using edge/fog computing. Merging
these different processes, inputs and outputs in a number of ways will result in a
plethora of resolutions as well as hundreds of operational and commercial models,
accelerating the Industrial 4.0 Revolution in post-­Covid era. This chapter provides
a comprehensive analysis to highlight the most recent advancements in fog comput-
ing activities in healthcare related to the IoT. Other aspects of the edge/fog com-
puting area, such as application support styles, methodologies and resources, are
274  Intelligent network design driven by Big Data analytics

also discussed. Finally, the importance of edge/fog computing in the context of the
Covid-­19 pandemic is discussed. Future research should concentrate on improving
present edge tools, and there is a pressing necessity for the development of a strong
computationally edge intelligent devices and model for tackling pandemic.

References

[1] Wu Q., He K., Chen X. ‘Personalized federated learning for intelligent IoT
applications: a cloud-­edge based framework’. IEEE computer graphics and
applications. 2020, vol. 1, pp. 35–44.
[2] Stoyanova M., Nikoloudakis Y., Panagiotakis S., Pallis E., Markakis E.K. ‘A
survey on the Internet of things (IoT) forensics: challenges, approaches, and
open issues’. IEEE Communications Surveys & Tutorials. 2020, vol. 22(2),
pp. 1191–221.
[3] Abrahão M.T.F., Nobre M.R.C., Gutierrez M.A. ‘A method for cohort se-
lection of cardiovascular disease records from an electronic health record
system’. International journal of medical informatics. 2017, vol. 102, pp.
138–49.
[4] Agniel D., Kohane I.S., Weber G.M. ‘Biases in electronic health record data
due to processes within the healthcare system: retrospective observational
study’. BMJ. 2018, vol. 361, p. k1479.
[5] Baumann L.A., Baker J., Elshaug A.G. ‘The impact of electronic health re-
cord systems on clinical documentation times: a systematic review’. Health
Policy. 2018, vol. 122(8), pp. 827–36.
[6] Tiwari S. ‘An ensemble deep neural network model for onion-­routed traffic
detection to boost cloud security’. International Journal of Grid and High
Performance Computing. 2021, vol. 13(1), pp. 1–17.
[7] Green D.R. Cloud Computing in Healthcare: Understanding User Perception,
Organizational Operations, and IT Costs to Be Successful in the Cloud.
[Doctoral dissertation]. Northcentral University; 2020.
[8] Mourya A.K., Idrees S.M. ‘Cloud computing-­based approach for accessing
electronic health record for healthcare sector’. Microservices in Big Data
Analytics. Singapore: Springer; 2020. pp. 179–88.
[9] Dong P., Ning Z., Obaidat M.S., et al. ‘Edge computing based healthcare sys-
tems: enabling decentralized health monitoring in Internet of medical things’.
IEEE Network. 2020, vol. 34(5), pp. 254–61.
[10] Amin S.U., Hossain M.S. ‘Edge intelligence and Internet of things in health-
care: a survey’. IEEE Access. 2020, vol. 9, pp. 45–59.
[11] Patra B., Mohapatra K. ‘Cloud, edge and fog computing in healthcare’.
Intelligent and Cloud Computing. Singapore: Springer; 2021. pp. 553–64.
[12] Verma P., Fatima S. 'Smart healthcare applications and real-­time analytics
through edge computing' in Internet of things use cases for the healthcare
industry; 2020. pp. 241–70.
 A review of edge computing in healthcare internet of things  275

[13] Yaraziz M.S., Bolhasani H. 'Edge computing applications for IoT in health-
care: A systematic literature review'. In Review. 2021.
[14] Haghi Kashani M., Madanipour M., Nikravan M., Asghari P., Mahdipour
E. ‘A systematic review of IoT in healthcare: applications, techniques, and
trends’. Journal of Network and Computer Applications. 2021, vol. 192(5)
103164.
[15] Dang L.M., Piran M.J., Han D., Min K., Moon H. ‘A survey on internet of
things and cloud computing for healthcare’. Electronics. 2019, vol. 8(7),
p. 768.
[16] Singh A., Chatterjee K. ‘Securing smart healthcare system with edge comput-
ing’. Computers & Security. 2021, vol. 108(1) 102353.
[17] Mulimani M.S., Rachh R.R. ‘Edge computing in healthcare systems’. Deep
Learning and Edge Computing Solutions for High Performance Computing.
Cham: Springer; 2021. pp. 63–100.
[18] Qayyum A., Qadir J., Bilal M., Al-­Fuqaha A. ‘Secure and robust machine
learning for healthcare: a survey’. IEEE Reviews in Biomedical Engineering.
2020, vol. 14, pp. 156–80.
[19] Saleem T.J., Chishti M.A. ‘Exploring the applications of machine learning in
healthcare’. International Journal of Sensors, Wireless Communications and
Control. 2020, vol. 10(4), pp. 458–72.
[20] Tiwari S. ‘Dermatoscopy using multi-­layer perceptron, convolution neural
network, and capsule network to differentiate malignant melanoma from be-
nign nevus’. International Journal of Healthcare Information Systems and
Informatics. 2021, vol. 16(3), pp. 58–73.
[21] Sharma S., Tiwari S. ‘COVID-­19 diagnosis using X-­ray images and deep
learning’. International Conference on Artificial Intelligence and Smart
Systems (ICAIS); Coimbatore, India, IEEE, 2021. pp. 344–49.
[22] Darwish A., Hassanien A.E., Elhoseny M., Sangaiah A.K., Muhammad K.
‘The impact of the hybrid platform of Internet of things and cloud computing
on healthcare systems: opportunities, challenges, and open problems’. Journal
of Ambient Intelligence and Humanized Computing. 2019, vol. 10(10), pp.
4151–66.
[23] Ning H., Li Y., Shi F., Yang L.T. ‘Heterogeneous edge computing open plat-
forms and tools for Internet of things’. Future Generation Computer Systems.
2020, vol. 106, pp. 67–76.
[24] Ranjan R., Villari M., Shen H., Rana O., Buyya R. ‘Software tools and tech-
niques for FOG and edge computing’. Software: Practice and Experience.
2020, vol. 50(5) 473–5.
[25] Gupta H., Vahid Dastjerdi A., Ghosh S.K., Buyya R. ‘iFogSim: a toolkit for
modeling and simulation of resource management techniques in the Internet
of things, edge and FOG computing environments’. Software: Practice and
Experience. 2017, vol. 47(9), pp. 1275–96.
[26] Javaid M., Khan I.H. ‘Internet of things (IoT) enabled healthcare helps to
take the challenges of COVID-­19 pandemic’. Journal of Oral Biology and
Craniofacial Research. 2021, vol. 11(2), pp. 209–14.
276  Intelligent network design driven by Big Data analytics

[27] Sufian A., Ghosh A., Sadiq A.S., Smarandache F. ‘A survey on deep transfer
learning to edge computing for mitigating the COVID-­19 pandemic’. Journal
of Systems Architecture. 2020, vol. 108(4), p. 101830.
[28] Rahman M.A., Hossain M.S. ‘An internet-­of-­medical-­things-­enabled edge
computing framework for tackling COVID-­19’. IEEE Internet of Things
Journal. 2021, vol. 8(21) 15847–54.
[29] Tiwari S., Jain A. ‘Convolutional capsule network for COVID‐19 detection
using radiography images’. International Journal of Imaging Systems and
Technology. 2021, vol. 31(2), pp. 525–39.
[30] Kong X., Wang K., Wang S., et al. ‘Real-­time mask identification for
COVID-­19: an edge-­computing-­based deep learning framework’. IEEE
Internet of Things Journal. 2021, vol. 8(21) 15929–38.
[31] Varghese B., Wang N., Barbhuiya S., Kilpatrick P., Nikolopoulos D.S.
‘Challenges and opportunities in edge computing’. IEEE International
Conference on Smart Cloud (SmartCloud); New York, NY, USA, IEEE, 2016.
pp. 20–26.
[32] Xiao Y., Jia Y., Liu C., Cheng X., Yu J., Lv W. ‘Edge computing security: state
of the art and challenges’. Proceedings of the IEEE. 2019, vol. 107(8), pp.
1608–31.
[33] Liu S., Liu L., Tang J., Yu B., Wang Y., Shi W. ‘Edge computing for autono-
mous driving: opportunities and challenges’. Proceedings of the IEEE. 2019,
vol. 107(8), pp. 1697–716.
[34] Haidar M., Kumar S. ‘Smart healthcare system for biomedical and health care
applications using aadhaar and blockchain’. 5th International Conference
on Information Systems and Computer Networks (ISCON); Mathura, India,
IEEE, 2022. pp. 1–5.
[35] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[36] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[37] Punhani A., Faujdar N., Kumar S. 'Design and evaluation of cubic to-
rus network-­on-­chip architecture'. International Journal of Innovative
Technology and Exploring Engineering (IJITEE), 1672-­1676. 2019, vol.
8(6).
[38] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based Sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[39] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
 A review of edge computing in healthcare internet of things  277

[40] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road Surface Quality Monitoring
Using Machine Learning Algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy Sustainability.
Smart Innovation, Systems and Technologies. 265. Singapore: Springer; 2022.
[41] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and iot for smart
road traffic management system’. Proceedings of the IEEE India Council
International Subsections Conference, INDISCON 2020, Publisher: IEEE,
Visakhapatnam, India; 2020. pp. 289–96.
[42] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementa-
tion of fault tolerance technique for internet of things (iot)’. Proceedings
– 2020 12th International Conference on Computational Intelligence and
Communication Networks, CICN 2020, Publisher: IEEE, Bhimtal, India;
2020. pp. 154–59.
[43] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. Proceedings of the
Confluence 2020 – 10th International Conference on Cloud Computing, Data
Science and Engineering; Noida, India, IEEE, 2020. pp. 63–76.
[44] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); India, Mathura, 2021. pp. 1–6.
[45] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 4th International Conference on Information Systems and
Computer Networks, ISCON 2019; Mathura, India, IEEE, 2019. pp. 250–55.
[46] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
Proceedings of the International Conference on Computational Intelligence
and Communication Networks, CICN 2015; 2016. pp. 79–84.
[47] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to mac layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium; The Electromagnetics Academy,
2015. pp. 2363–67.
[48] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wire-
less sensor network’. Proceedings of the 5th International Conference on
Communication Systems and Network Technologies, CSNT; Gwalior, India,
IEEE, 2015. pp. 194–200.
[49] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, IEEE, 2014.
[50] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based trans-
parent and secure decentralized algorithm’. International Conference on
Intelligent Computing and Smart Communication. Algorithms for Intelligent
Systems; Springer: Singapore; 2020.
278  Intelligent network design driven by Big Data analytics

[51] Kumar S., Trivedi M.C., Ranjan P., Punhani A. Evolution of Software-­Defined
Networking Foundations for IoT and 5G Mobile Networks. Hershey, PA: IGI
Publisher; 2020. p. 350.
[52] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
Personal Communications. 2021, vol. 10(3).
 Chapter 13
Image Processing for medical images on the
basis of intelligence and biocomputing
M. Mohammed Mustafa 1, S. Umamaheswari 2, and
Korhen Cengiz 3

Intelligence in medical imaging explores how intelligent computing can create a

large amount of changes to existing technology in the field of medical image pro-
cessing. The book presents various algorithms, techniques, and models for inte-
grating medical image processing with artificial intelligence (AI) and biocomput-
ing. Bioinformatics solutions lead to an effective method for processing the image
data for the purpose of retrieving the information of interest and collecting various
data sources for extracting the knowledge. Moreover, image processing methods
and techniques help scientists and physicians in the medical field with diagnosis
and therapies. It describes evolutionary optimization techniques, support vector
machines (SVMs), fuzzy logic, a Bayesian probabilistic framework, a reinforcement
learning-­based multistage image segmentation algorithm, and a machine learning
(ML) approach. It discusses how these techniques are used for image classifica-
tion, image formation, image visualization, image analysis, image management, and
image enhancement. The term “medical image processing” illustrates the provision
of digital image processing, particularly for medicine. Medical imaging intends to
identify internal structures hidden in the human body. It helps to find abnormalities
in the body. Digital images can be processed effectively, also evaluated, and utilized
in many circumstances concurrently with help of suitable communication protocols.

13.1 Introduction

The motive for the primary image processing was modified to enhance the huge
image. It is modified to goal human beings to enhance the seen effect of people. In

1
Department of Information Technology, Sri Krishna College of Engineering & Technology,
Coimbatore, Tamilnadu, India
2
Department of Information Technology, C. Abdul Hakeem College of Engineering & Technology,
Melvishsaram, Tamilnadu, India
3
College of Information Technology, University of Fujairah, UAE
280  Intelligent network design driven by Big Data analytics

image processing, the middle is a low and huge image, and the output is an image
with top-­notch progress. Common image processing consists of image enhance-
ment, restoration, encoding, and compression. First-­a-­hit software program became
the American Jet Propulsion Laboratory. They used image processing techniques,
which protected geometric correction, gradation transformation, noise removal, etc.

13.1.1 What is an image?
Practically, each scene around us is bureaucratic, and this is concerned with photo
processing. A photo is shaped through two-­dimensional analog and virtual signs that
carry color facts organized alongside x and y spatial axis.

13.2 Image processing

Image processing is a technique to carry out a few operations on an image, with the
purpose to get a superior image or extract a few beneficial records from it.
In general terms, manipulating an image to amplify the same to induce informa-
tion out of it is called image processing.
There are two styles of image processing, which are as follows:

• analog image processing, which is used for recycling photos, printouts, and
other image hard clones
• digital image processing, which is used for manipulating digital images with the
help of complex algorithms

The main purposes of image processing are as follows:

• representing reused data in a visual way one can understand, for example, giv-
ing a visual form to unnoticeable objects
• to ameliorate the reused image quality, image stropping and restoration work
well
• image recuperation helps in searching images
• helps to measure objects in the image
• with pattern recognition, it becomes easy to classify objects in the image, detect
their position, and get an overall understanding of the scene

13.2.1  Equivalent image processing

Analog image processing is carried out on analog alerts, and its methods are the sim-
plest two-­dimensional alerts. The snapshots are manipulated with the aid of using
electric alerts. In analog image processing, analog alerts may be periodic or nonpe-
riodic. Examples of analog snapshots are TV snapshots, images, paintings, clinical
snapshots, etc. [1].
 Image Processing for medical images  281

13.2.2  Digital image processing

Virtual image processing is carried out on virtual images (a matrix of small
pixels and elements). For manipulating the images, there are some software
programs and algorithms, which are carried out to carry out changes. Digital
image processing is one of the quickest developing enterprises which impacts
everyone’s life. Examples of virtual images are processing, image recognition,
video processing, etc.

13.2.3  Digital image

A virtual image is an illustration of a two-­dimensional image as a fixed set of virtual
values, referred to as image factors or pixels.
Common image formats are as follows:

• one sample per point (grayscale or B&W)

• two samples per point (green, red, and blue)
• three samples per point (green, blue, red, and “Alpha”)

13.2.4  Applications of color models

Red, Green, Blue (RGB) model for color monitor and color video camera CMY
mode included for color printing HIS mode included for color image processing
YIQ mode included for color image transmission.
Digital photo processing represents the usage of a personal computer (PC) to govern
images through an algorithm. Analog photo processing allows a far wider form of algo-
rithms to be applied to enter statistics and might keep away from issues together with
noise and distortion at some point of processing. Since images are defined in phrases
of dimensions, digital photo processing may be modeled in the shape of a multidimen-
sional system. Screening of huge numbers of people is the want of the hour to minimize
the unfolding of ailment within the community. Real-­time Polymerase chain reaction
(PCR) is a widespread diagnostic device getting used for pathological checking out.
But the growing wide variety of fake check effects has opened the route for exploration
of opportunities for checking out tools. Chest X-­rays of Covid-19 patients affected by
pneumonia, tuberculosis, and chronic bronchitis have proved to be a crucial opportunity
indicator in disease screening. But again, accuracy relies upon radiological expertise [2].

13.2.5  Applications of digital image processing

The use of virtual photograph manner strategies has exploded, and that they are
presently used for various duties altogether in variety of regions, such as image
enhancement and restoration, medical field, remote sensing, transmission and cod-
ing, machine/robot vision, color processing, pattern recognition, video processing,
and microscopic images.
282  Intelligent network design driven by Big Data analytics

13.2.6  Fundamental steps in digital image processing

Step 1: Image acquisition
The photograph is captured via means of a device (e.g., camera) and digitized if
the output of the digital or sensor isn’t perpetually in virtual form, using the analog-­
to-­virtual device [3–8].
Step 2: Image enhancement
The method of manipulating an image in order that the end result is greater
appropriate than the unique for particular applications. The concept behind enhance-
ment strategies is to convey information that might be hidden, or easy to spotlight
the positive functions of a hobby in an image.
Step 3: Image restoration
It improves the arrival of associate degree image – tend to be mathematical or
probabilistic models. Enhancement, on the choice hand, is entirely on human sub-
jective selections regarding what constitutes a “good” sweetening result.
Step 4: Color image processing
Use the shade of the image to extract the capabilities of the hobby in a very
image. Color modeling and process in a virtual space plays an important role in
image processing.
Step 5: Wavelets
These are the muse of representing photos in various stages of resolution. It is
used for image facts compression whereby photos are divided into smaller regions.
Step 6: Compression techniques
These are used for decreasing the garage required to keep an image or the band-
width required to transmit it.
Step 7: Morphological processing
Tools for extracting photographic additives that can be useful in the illus-
tration and outline of the shape. In this step, there may be a transition from
approaches that generate images, to approaches that emit photographic attributes
[9–12].
Step 8: Image segmentation
Segmentation approaches break down an image into its building blocks or
objects. Important tip: The better the segmentation, the more likely the popularity
to succeed.
Step 9: Representation and description
Representation: select whether or not the records should be represented as a
boundary or as an entire region. It continuously follows the exit from a segmentation
phase. Representation of limits: focus on the characteristics of the external form, as
well as angles and inflections. Representation of the region: focus on internal prop-
erties, as well as the texture or shape of the skeleton. Transform raw recordings into
a suitable form for further processing on PC. Descriptive offerings with extractive
attributes that determine certain quantitative hobby records are fundamental to dif-
ferentiate elegance of one object from another.
Step 10: Object recognition
 Image Processing for medical images  283

The method of labeling an item, based primarily on the facts, is provided by its
description. Recognition is the method of assigning a tag, such as a “vehicle” to an
item, based primarily on its descriptors.

13.2.7  Components of an image processing system

1. Image sensors

The first is the bodily instrument, which is sensitive to the electricity radi-
ated using the object that one wishes to visualize (sensor). The second,
called a digitizer, is a tool for transforming the output of the body detection
tool into a virtual shape [13–19].

2. Specialized image processing hardware

It typically includes the digitizer, mentioned earlier, as well as hardware

that performs several primitive operations, including a mathematical
good judgment unit (ALU), which reproduces mathematical and logi-
cal operations in parallel to the entire images. This form of hardware
is sometimes known as the front-­end subsystem, and its biggest distin-
guishing feature is speed. In other words, this unit performs features that
require statistical processing speeds that the standard laptop computer
cannot handle.

3. Image processing software

Image processing software includes specialized modules that perform

unique tasks. A well-­designed package also includes the functionality to al-
low the person to write code that, at a minimum, uses specialized modules.

4. Mass storage capability

Mass garage functionality is a necessity in an image processing program.

And an image of size 1,024 × 1,024 pixels requires a megabyte of garage
area if the image is not always compressed. The digital garage for image
processing programs falls into three main categories: short-­term archiving
for use during processing, online archiving for relatively quick recall, and
archiving characterized by infrequent access.

5. Image displays

Presentations used today are mainly color television screens (ideally a flat
screen). Monitors are pushed through the image outputs, and the images
show playing cards which are an essential part of a PC system.
284  Intelligent network design driven by Big Data analytics

6. Hardcopy devices

Used for recording images, they consist of laser printers, film cameras,
heat-­sensitive devices, inkjet gadgets, and virtual gadgets, which include
optical discs, cameras, and Compact Disk-­Read Only Memory (CD-­ROM).

7. Networking

It is a kind of default characteristic on any transportable device in use to-

day. Due to the large quantity of data inherent in photograph processing
software program packages, the essential aim of photograph transmission
is bandwidth. In devoted networks, this isn’t always typically usually a
problem; however, communications with far-­flung websites in the com-
munity aren’t usually as efficient.

13.3 Medical imaging

Medical imaging is the technique used to achieve pictures of the frame elements for
scientific applications, makes use of it on the way to perceive or examine diseases.
There are masses of heaps of imaging strategies completed every week worldwide.
Medical imaging is developing swiftly due to developments in image graph process-
ing techniques including image graph recognition, assessment, and enhancement.
Image processing will grow the percentage and amount of detected tissues. There
are many straightforward and complicated image graph assessment techniques
within the medical imaging field. They are also explaining a way to exemplify photo
interpretation demanding situations and the usage of distinct photo processing algo-
rithms which include k-­means, ROI primarily based on segmentation, and water-
shed strategies.

13.4 Deep learning techniques

Deep learning strategies, especially convolutional neural networks (CNNs), have con-
firmed success in clinical imaging classification. Many, one-­of-­a-­kind CNN architectures
have been investigated on images of chest X-­rays for prognosis of sickness (Figure 13.1).
These fashions were preskilled at the keras and tensor waft ImageNet database with the aid
of lowering the use for big education units as they have got preskilled weights. Its changes
discovered that CNN primarily based on architectures has the capability for prognosis
of sickness. Several types of current in-­depth study strategies including Convolutional
Neural Community (CNN), Vanilla Neural Community, Fully Neural Community-­Based
Visible Geometry Organization (VGG), and Pill Community are implemented for the pre-
diction of lung disease. Simple CNN has terrible overall performance for rotated, tilted,
or non-­normal image orientation. Recently, in-­depth study has proven brilliant capabil-
ity while implementing clinical images for sickness detection, inclusive of lung sickness.
 Image Processing for medical images  285

Figure 13.1    The architecture of image processing using deep learning

Lung sickness refers to problems that have an effect on the lungs, the organs that enable
us to breathe. Breathing trouble due to lung sickness may also save the frame from getting
sufficient oxygen. Lung sickness is the first-­rate challenge for human beings, and many
people lose their life due to lung sickness. In general, the CNN set of rules is getting used
for image classification and popularity due to its excessive accuracy.
In this modern day ever transcending world, wherein the generation is convert-
ing loads of one-­of-­a-­kind fields and industries, human beings nevertheless now
no longer create an entire benefit of generation in diverse industries. The foremost
cause for this will be that maximum of the engineers as a minimum the bulk of
them are absolutely dedicated to laptop technological know-­how, which regard a
sequence of hate or problem attitude for the programmer; however, there are few
who clearly desire to make large adjustments in the clinical region and in fact make
novel improvements within the field.
In current years, a lot of such computer-­aided diagnoses (CAD) structures are
designed for the analysis of numerous diseases. Most lung cancer detection at an early
degree has come to be very critical and additionally very clean with image processing and
deep gaining knowledge of strategies. The lung-­affected person computer tomography
(CT) test images are used to classify the lung nodules and to stumble on the malignancy
stage of those nodules. The CT test images may be segmented with the use of U-­Net archi-
tecture. Globally, pneumonia is the maximum critical motive of loss of life, although it
286  Intelligent network design driven by Big Data analytics

is a vaccine-­preventable ailment. It may be detected by studying chest X-­rays. Analyzing

chest X-­rays is a hard assignment and calls for precision. A higher and superior synthetic
intelligence device for pneumonia detection can move in an extended manner to lowering
the mortality price and growing lifestyle expectancy. A deep neural community version
primarily based on CNNs and residual community together with strategies of figuring out
the most reliable differential quotes, the use of cosine annealing and stochastic gradient
with restarts to reap a green, and enormously correct community with a view to assisting
stumble on and are expecting the presence of pneumonia with the use of chest X-­rays.
Lung disease is a common occurrence at some point in the world. These
embrace persistent clogging pneumonic upset, pneumonia, asthma, tuberculosis,
fibrosis, etc. Timely analysis of respiratory organ ailment is essential. Many image
processing and gadget gaining knowledge of fashions had been evolved for this pur-
pose. Millions of human beings have developed coronary heart ailment each year,
and coronary heart ailment is the most important cause of fatality of both women
and men in the United States and across the world. The World Health Organization
(WHO) analyzed that 12 million deaths take place globally because of heart dis-
eases. In nearly every 34 seconds, the coronary heart ailment kills one person in the
world. Medical analysis performs critical functions and complex assignments that
require to be done efficaciously and accurately. To lessen fee for attaining medical
exams, the appropriate PC primarily based on statistics and choice guide need to be
aided. Data mining is using software program strategies for locating styles and con-
sistency in units of records. Also, with the appearance of records mining during the
final decades, there is a massive possibility to permit computer systems to at once
assemble and classify the specific attributes or classes.
A deep neural network model based primarily on CNNs and residual networks
collectively with reliable maximum differential rate identification techniques using
cosine annealing and stochastic gradient with restarts to obtain an inexperienced
and distinctly precise network that helps to meet and anticipate the presence of
pneumonia in chest X-­rays. Lung disorder is not an unusual place at an unspecified
time in the future of the world. These include continuous obstructive pulmonary
disorders, pneumonia, asthma, tuberculosis, fibrosis, etc. Rapid assessment of lung
disorders is essential. Many models of image processing and machine mastering
have been advanced for this purpose. Millions of humans have emerged with some
shape of coronary heart disorder every year, and coronary heart disorder is the
maximum cause of death of both women and men in the United States and through-
out the world. The WHO analyzed that 12 million deaths take place globally due
to heart diseases. In almost every 34 seconds, the coronary heart disorder kills
one man or woman in the world. Medical evaluation plays a vital characteristic;
however, complicated project wants to be achieved successfully and accurately.
To reduce rate for achieving scientific checks, an appropriate laptop-­based definite
information and desire manual need to be aided. Data mining is the use of software
program application techniques for finding patterns and consistency in devices of
data. Also, with the advent of data mining during the final decades, there is a big
opportunity to allow PC structures to gather and classify the unique attributes or
classes.
 Image Processing for medical images  287

13.4.1  Uses of image processing

1. Image correction, stropping, and resolution Correction

Frequently, we wish we could make old images more. And that is possible
currently. Zooming, stropping, edge discovery, and high dynamic range
edits all fall under this order. All these ways help in enhancing the image.
Utmost editing software and image correction law can do these effects flu-
ently.

2. Pollutants on editing apps and social media

Utmost editing apps and social media apps give pollutants these days.

Pollutants make the image look more visually charming. Pollutants are
generally a set of functions that change the colors and other aspects in an
image that make the image look different. Pollutants are an intriguing op-
eration of image processing.

3. Medical technology

In the medical field, image processing is used for colorful tasks such as
Positron emmission tomography (PET) checkup, X-­ray imaging, medical
CT, UV imaging, cancer cell image processing, etc. The preface of image
processing to the medical technology field has greatly bettered the diag-
nostic process.

4. Computer/machine vision

One of the most intriguing and useful operations of image processing is in

the computer vision. Computer vision is used to make the computer see,
identify effects, and process the whole terrain as a whole. The important
use of computer vision is self-­driving buses, drones, etc. Computer vision
helps in handicap discovery, path recognition, and understanding the ter-
rain.

5. Pattern recognition

Pattern recognition is a part of image processing that involves AI and ma-

chine literacy. Image processing is used to find out colorful patterns and
aspects in images. Pattern recognition is used for handwriting analysis,
image recognition, computer-­backed medical opinion, etc.

6. Video processing
288  Intelligent network design driven by Big Data analytics

Video is principally a fast movement of images. Colorful image processing

ways are used in video processing. Some styles of video processing are
noise junking, image stabilization, frame rate conversion, detail improve-
ment, etc.

13.5 Convolutional neural network

A CNN is crafted from one or more convolutional layers (regularly with a subsampling
step) as soon as that is observed through technique of manner of one or more wholly
related layers as accomplice diploma exceeding in a powerful multilayer neural commu-
nity. The shape of a CNN is supposed to require a gain of the second one shape of an input
image (or absolutely distinct two-­dimensional input which includes a speech signal). This
may be completed with near connections and tied weights found through the manner of
many forms of pooling which ends up in translation-­invariant features. Another gain of
CNNs is that they are simpler to train and feature numerous fewer parameters than wholly
related networks with the same form of hidden units. The data are preprocessed which
include image reshaping, resizing, and conversion to array form. A similar system is also
accomplished on the test image. A database complete of approximately 4,000 one every
of a kind plant species is obtained, out of that, any photograph is likewise used as a test
photograph for the software bundle program.
The training database is used to teach the version (CNN) so that it can select the con-
trol image and the disease it has. CNN has unique levels which could be Convolution2D,
MaxPooling, and fully related. Once the version is successfully activated, the software can
detect the disease if the plant species is contained in the database.
After a fulfillment training and preprocessing, evaluation takes a look at image,
and skilled model takes area to anticipate the contamination.

1. This model will be a successive model having a chain of layers to transform

equal preceding period image proper right into a carry out set for similar
processing.
2. The number one layers of this model M are the convolutional layer with 128
filters, and Relook because of the activation feature.
3. The 0.33 layer is that the Gamma hydroxybutrayte (GHB) pooling layer, so it
is going to lessen the dimensions of the convoluted image through manner of
(2, 2).
4. It yet again offers in addition convolutional layers can have 256 filters and recti-
fied linear unit (ReLU) because of the activation feature.
5. The sixth layer is that the GHB pooling layer with a pool length of (2, 2).
6. A convolutional layer with 384 filters and ReLU activation feature is delivered
to the version.
7. The following layer is once more a convolutional layer with 256 filters and
ReLU activation feature.
8. Then it is a max pooling layer ascertained via means of dropout fee 0.2.
 Image Processing for medical images  289

9. Then this version has brought one further layer to flatten the output of the on top
of designed convolutional neural community version.
10. This destruction system will deliver the perform set for every image within the
form of output.
11. Now, this version has utterly connected layers so it will be used for the kind of
photos entirely at the generated function set.
12. This dense layer acts as a result of the hidden layer of the artificial neural com-
munity having 512 hidden neurons, and also the activation feature is ReLU.
This version is an intended one among these manners that every entered somatic
cell is connected to every completely different hidden neuron forming a very
connected layer.
13. It is one further completely related layer that acts as a result of the output layer
of the artificial neural community having three output neurons. The range of
output neurons is regularly looking on the classes. It makes use of SoftMax
activation feature.
14. The output of this residue is that the anticipated magnificence label that is to
assess the final accuracy of this version.

The CNN algorithm is being used for image classification and recognition
because of its high accuracy. Initially, the problem statement has been studied and
analyzed the dataset, then apply machine learning and deep learning to predict
whether the patient has lung disease or not. In this current ever-­transcending world,
where technology is changing a lot of different fields and industries, people still not
making complete advantage of technology in various industries. The main reason
for this would be that most of the engineers are completely committed to computer
science, which regard a series of hate or difficulty mindset for the programmer, but
there are few who really want to make big changes in the medical sector and actually
make novel advancements in the field.

13.6 Convolution layers
From Figure 13.2, the convolutional layer includes a tough associate in nursing
quick of filters whose parameters need to be learned. The height and weight of the
filters are smaller than those of the enter extent. Every filter is convolved with the
enter extent to cipher an activation map factory made from neurons. In numerous
words, the clear out is softened throughout the breadth and peak of the middle, and
also the dot merchandise between the enter and clear out is computed at each spatial
position.
Automatically detecting diseases and obtaining correct diagnoses through X-­ray
medical images have become a new research priority in the field of computer sci-
ence and AI, as the expense of manual labeling and classification continues to rise.
However, the quality of a standard radiograph is insufficient for most activities,
and traditional approaches are inadequate for dealing with large pictures. To detect
pneumothorax from chest X-­ray images, we present a feature fusion CNN model. To
290  Intelligent network design driven by Big Data analytics

Figure 13.2    The structure of the convolutional layer

begin, two methods are used to improve the preprocessed image samples. The final
classification is then implemented using a feature fusion CNN model that combines
the Gabor features with the additional information collected from the images.
Furthermore, the CNN has made significant progress in the recognition of pat-
terns in images, particularly in medical imaging. Designing an image-­feature extrac-
tor is crucial in traditional CAD techniques. This is, however, a challenging task.
A CAD approach that employs CNN, on the other hand, does not necessitate the
employment of an image-­feature extractor. We used CNN to create an image-­based
Computer Aided Design X (CADx) for differential diagnosis of lung anomalies
including lung nodules and diffuse lung illnesses in this work. CNN performs admi-
rably in the classification of natural photos. As a result, numerous studies have been
conducted on the differential diagnosis of lung anomalies such as lung nodules and
diffuse lung illnesses. CNNs or ConvNets are a type of deep neural network used to
analyze visual imagery in deep learning. Based on the shared weight architecture of
the convolution kernels that shift over input features and produce translation equiv-
ariant responses, they are also known as shift in variant or space invariant artificial
neural networks (SIANN). Surprisingly, most convolutional neural networks are
equivariant rather than invariant under translation. They can be used in computer
vision and pattern recognition, recommender systems, classification, image segmen-
tation, medical image analysis, natural language processing, central nervous system
interfaces, and financial time series, to name a few applications.
Step 1: The first step is to do a convolution operation. The convolution opera-
tion is the initial step in our attack strategy. We will discuss about feature detectors
in this stage, which are essentially the neural network’s filters. We will also discuss
about feature maps, including how to learn the parameters of such maps, how to
recognize patterns, the layers of detection, and how to map out the results.
 Image Processing for medical images  291

Step 1a: ReLU layer will be used in the second half of this process. We will
discuss about ReLU layers and how linearity works in the context of CNNs. It is not
vital to understand CNNs but it is never a bad idea to brush up on your skills.
Step 2: Pooling is the next step. Pooling will be covered in this section, and we
will learn how it works in general. However, our nexus will be a special form of
pooling: maximum pooling. However, we will go over a variety of ways, including
mean (or sum) pooling. This section will conclude with a demonstration using a
visual interactive tool that will undoubtedly clarify the entire subject.
Step 3: Flattening is the third step. When working with CNNs, there will be a
brief discussion of the flattening process and how we move from pooling to flattened
layers.
Step 4: Complete the connection, and everything we have discussed so far will
be combined in this section. By learning this, you will gain a better understanding of
how CNNs work and how the “neurons” that are eventually formed learn to classify
photos.
The output extent of the convolutional layer is received via manner of means of
stacking the activation maps of all filters aboard the intensity dimension. Since the
breadth and peak of each filter out are intended to be smaller than the entire, every
nerve cell within the activation map is simplest involving tiny low close space of
the enter extent. In numerous words, the receptive subject length of every neuron
is small and is a clone of the clear out length. The nearby property is influenced via
way of means of the structure of the animal visible cortex whereby the receptive
fields of the cells are small. The nearby property of the convolutional layer permits
the community to check filters that maximally reply to a close-­by space of the enter,
consequently exploiting the spatial nearby correlation of the enter (for AN enter
image, a constituent is further correlated to the within sight pixels than to the remote
pixels). In addition, as a result, the activation map is received via manner of means
of acting convolution among the filter out and also the larger, the clear out param-
eters are shared for all nearby positions. The load sharing reduces the large choice
of parameters for performance of expression, performance of learning, and precise
generalization.
The accuracy of the version is excessive while compared with different category
methods.

1. The version is computationally powerful and simple.

2. Due to actual time implementation, it will increase financial value.

After categorizing the photos based mostly on elegance labels, process the facts
to fit in our model with the usage of CNN and educate the CNN based mostly on
training photos. CNN is proposed on this artwork for the elegance of the leaves
inflamed with the brilliant sorts of fungal disease. The education snapshots have
been taken for the beauty labels. Training a CNN is the exercising of going for
walk education examples through the version from the input layer to the output
layer concurrently developing a prediction and identifying the effects or errors. If
292  Intelligent network design driven by Big Data analytics

the prediction is incorrect then it is over again propagated in opposite order, that is,
from final layer to first layer.

13.6.1  Training phase

During the training segment, the version’s inner weights are mechanically up to date
over numerous iterations. External elements including the training method, struc-
ture, regularization techniques, or the rate of the hyperparameters have an effect on
this training process. Comparing research and their consequences to extract insights
on the way to outline the training segment is complex due to the reality that they do
not use the identical records and do not offer all of the parameters required to breed
their experiments.

13.6.2  Training strategies

There are techniques to educate a CNN: from scratch or with switch studying.
Transfer studying is at the same time as a community is preinformed on a massive
set of photos (e.g., ImageNet and its 1.2 million photos in 1,000 classes) is used and
tailored to three distinctive tasks. This shape of studying is enabled through way
of method of the fact that the primary layers of CNNs take a look at common low-­
diploma functions that aren’t splendor unique.
In practice, this model is finished using the community weights from preced-
ing education. Using switch analyzing permits us to apply CNNs even though the
quantity of education data is limited, that is frequently the case within the context
of crop illnesses identification. This approach allows to benefit more generalizabil-
ity, because of the truth that the community had formerly located to address lot of
examples. It is likewise a manner to keep in phrases of computing time and capac-
ity. There are tactics to carry out switch analyzing: via feature extraction and fine-­
tuning. Feature extraction is composed in maintaining the weights of a preinformed
version intact, and using the embedding it produces to educate a modern classifier
at the cause dataset. Fine-­tuning is composed with in the usage of the weights of
a preinformed version to initialize the version after which education all or a part
of those weights at the cause dataset. Choosing one approach or the opportunity is
primarily based upon the proximity among each of the supply and cause datasets (in
case they are very close, feature extraction can be sufficient), however, on the scale
of the cause dataset. Training a big quantity of layers with a small dataset can also
furthermore boom the danger of over becoming. Training from scratch is at the same
time as the community weights aren’t inherited from a preceding version, however,
are as a substitute randomly initialized. It calls for a bigger education set, and the
over becoming danger is better for the reason that community has no revel in preced-
ing education periods and so needs to depend upon the input data to outline all its
weights. However, this technique permits us to outline a problem-­particular commu-
nity form that may enhance the performance. These problem-­particular architectures
may be developed, as an example, to address more than three caseation channels,
multiscale dimensions, or to combine multiple fashions informed differently (with
diverse hyperparameters or datasets).
 Image Processing for medical images  293

13.6.3  CNN performance

A characteristic fusion convolutional neural community (CNN) version may be used
to locate pneumothorax from chest X-­ray photos. In this version, the preprocessed
image samples are stronger via means of strategies. Using CNN and regions with
CNN features (R-­CNN), the convolutional neural community (CNN) has delivered
approximately a step forward in sample reputation of photos which includes scien-
tific photos. In ordinary CAD algorithms, designing an image-­function extractor is
important. However, this mission is hard. On the other hand, a CAD set of guide-
lines through the manner of use of CNN does not usually require the image-­function
extractor. In this version, an image-­based CADx can be advanced for differential
diagnosis of lung abnormalities inclusive of lung nodules and diffuse lung ailments
through the way of the use of CNN. CNN shows immoderate general overall perfor-
mance for a class of natural photos. Therefore, many researchers have studied about
differential diagnosis of lung abnormalities including lung nodules and diffuse lung
ailments; on the other hand, a manner of R-­CNN became proposed for the detection
of devices including persons and animals in natural photos. Therefore, an image-­
based Computer assisted data entry (CADe) additionally may be used for the detec-
tion of lung abnormalities including lung nodules and diffuse lung ailments through
the way of the use of R-­CNN.

13.6.3.1 Interstitial lung disease detection

The usage of CNN in a large variety of lung texture styles of sickness may be
discovered in CT experiment photos. These snapshots are the intermix of several
patterns, and for that reason, it becomes very tough for radiologists to differenti-
ate them and diagnose the illness. One manner of fixing this hassle is the discov-
ery of CNN. CNN is usually used for sample magnificence and photo popular-
ity systems. They have completed an entire lot tons much fewer mistakes in the
database, and photo magnificence using CNN became exceedingly fast. Interstitial
lung infection is a time period that incorporates wonderful forms of lung infec-
tion. Insterstitial lung illnesses (ILDs) have an effect on the insterstitial, that is,
the part of the lung’s anatomic form. Lung tissue characterization is the critical
element of a CAD device for the detection of ILDs. Thus, using CNN, interstitial
lung infection detection offers the correct results. This version includes CNN hav-
ing seven layers with a local binary sample as a function extractor. The execution
of sophistication exhibited the capability of CNNs in studying lung patterns. In
our corpus, 15 research studies (79%) used switch studying and 7 research studies
(37%) informed a version from scratch. Choosing a schooling technique is primar-
ily based upon technical (quantity of photos, computing capacity) and thematic
(availability of the proper form or of preinformed weights properly best with the
records used) troubles in comparison with three schooling techniques on six CNN
architectures (AlexNet, DenseNet-­169, Inception v3, ResNet-­34, SqueezeNet-­1.1,
and VGG13). They used the Plant Village dataset augmented with the elegant facts.
Two techniques used switch studying: function extraction and whole fine-­tuning. In
one of the three techniques, the community came to be informed from scratch. The
294  Intelligent network design driven by Big Data analytics

accuracies were obtained at the validation set and the schooling time. For the six
architectures, fine-­tuning gave very brilliant precision (from 99.2% for SqueezeNet
to 99.5% for VGG13). The instances required for fine-­tuning and for schooling
from scratch are close (from 1.05 to 5, 64 h for fine-­tuning and from 1.05 to 5, 91
h on the equal time as informed from scratch). The function extraction technique
had the bottom schooling instances (from 0.85 to 3, 63 h). Overall, from scratch
and switch-­studying should recognize and must not be visible as without a doubt
together for the fantastic techniques.
Gao et al.’s artwork assign a single ILD class label without delay upon whole
axial CT test photographs but without preprocessing the reap ROIs. While reading
the lung ailments photo database, with segmentation masks, a variety of CT test
photographs are located with or more illness labels. The efficiency of deep gain-
ing knowledge is a subfield of device gaining knowledge regarding algorithms
stimulated via means of the feature and shape of the brain. Recent traits in device
gaining knowledge, in particular, aid the identity, quantification, and category of
styles in scientific photos. These traits had been made feasible because of the capac-
ity of deep gaining knowledge of two discovered capabilities simply from data, as
opposed to hand-­designed capabilities primarily based on domain-­precise knowl-
edge. Deep-­gaining knowledge is turning speedily into country of the art, mainly to
step forward overall performance in several scientific applications. Consequently,
those improvements help clinicians in detecting and classifying positive scientific
situations efficiently. Deep neural community fashions have conventionally been
designed, and experiments had been finished upon them by means of human spe-
cialists in a persevering manner with the trial-­and-­mistakes technique. This sys-
tem needs giant time, know-­how, and resources. To triumph over this problem, a
unique, however, easy version is brought to routinely carry out the most advanta-
geous category responsibilities with a deep neural community structure. The neural
community structure turned into especially designed for pneumonia image category
responsibilities. This approach is primarily based on the convolutional neural com-
munity set of rules, using a hard and fast set of neurons to convolve on a given image
and extract applicable capabilities from them. Demonstration of the efficacy of the
proposed approach with the minimization of the computational fee as the focal point
becomes done compared with the exiting modern lung illness class networks. Lung
infection detection is commonly offered by classifying an image graph into healthy
lungs or infection-­inflamed lungs. The lung infection classifier, every now and then
known as a version, is acquired through training. Training is the tool wherein a
neural community learns to apprehend a category of photos. Using deep getting
to know, it is miles feasible to train a version that can classify photos into their
respective beauty labels. Therefore, to use deep getting to know for lengthy infec-
tion detection, Step 1 is to collect photos of lungs with the infection to be classified.
The second step is to train the neural community till it can recognize the illnesses.
The very last step is to classify new photos. Here, new images are unseen through
the way of the version earlier than the established in the version, and the version
predicts the beauty of these photos.
 Image Processing for medical images  295

However, the process of deep learning applied to spot respiratory organ diseases
from medical images is described. There are three main steps: (1) image preprocess-
ing, (2) coaching, and (3) classification.

13.6.3.2 Image preprocessing
Image processing is a technique to carry out a few operations on an image, with a
purpose to get an additional image or to extract a few beneficial facts from it. It is
a form of sign processing wherein input is an image and output can be an image or
characteristics/competencies related to that image. Nowadays, photo processing is
among unexpectedly growing technologies. Image processing essentially consists of
the subsequent three steps:

1. importing the dataset image through image acquisition tools

2. analyzing and manipulating the dataset image
3. output wherein end result may be altered image or record, this is primarily based
on image analysis

13.6.3.3 Training
Deep mastering neural community fashions discover ways to map inputs to outputs
given an education dataset of examples. The education procedure entails locating
a fixed weight within the community that proves to be appropriate, or appropriate
enough, at fixing the unique problem. Training is the procedure wherein a neural
community learns to understand a category of images. We need to teach the neu-
ral community till it may understand the diseases. This is the vital section of the
machine as this section makes a decision on the accuracy and overall performance
of the machine.

13.6.3.4 Classification
The very last section is the type of the images. In our machine, the type is of the under-
neath types, such as (1) healthy, (2) pneumonia, (3) tuberculosis, and (4) Covid-­19. While
importing the image, our skilled version classifies the image into someone of the above
diseases. This approach has the subsequent advantages:

1. use higher and optimized facts of scans to supply an awful higher consequences
2. increase overall performance of utility computation usage
3. hyper rapid consequences without taking an awful computation from locale
4. accurate consequences
5. clear generalized record in my view dispatched to user
296  Intelligent network design driven by Big Data analytics

13.6.4  Convolutional neutral networks with AI

CNN can be a subfield of synthetic intelligence (AI). In deep learning, a CNN or
ConvNet is a category of deep neural networks, administrated to studying visual
imagery. They are also referred to as SIANN, due to their shared-­weights shape
and translation changelessness characteristics. There are several programs in image
and video recognition, recommender systems, image classification, clinical image
analysis, seasoned language processing, and financial time series. CNNs are regular-
ized variations of multilayer perceptron. Multilayer perceptions commonly counsel
fully connected networks, that is, each somatic cell in a single layer is hooked up
to all or any neurons within the next layer. The “absolute connectedness” of these
networks makes them in danger of over changing into records. Typical strategies
of regularization encompass some form of significance dimension of weights to
the loss function. CNNs take further ordinary unprecedented a rare methodology
toward regularization. They take gain of the hierarchical sample in records and col-
lect more difficult designs with the usage of smaller and fewer complicated styles.
Therefore, on the scale of connectedness and complexity, CNNs are at the decrease
extreme (Figure 13.3). Convolutional networks are aroused with the aid of exploita-
tion organic procedures, in this the property sample among neurons resembles the
leader of the animal visible cortex. Individual animal tissue neurons reply to stimuli
handiest in a very restricted space of the field of regard known as the receptive field.
The receptive fields of assorted neurons partially overlap such that they cowl the
entire visual view.

13.6.5  CNN layers

The CNN layer consists of a 32×32×3 image with a 5×5×3 filter and an activation
map that is used to slide over all the spatial locations.

Figure 13.3    The CNN layers with the activation map

 Image Processing for medical images  297

13.6.6  CNN image classifier

The classification of rubbish into recyclable and nonrecyclable is carried out by
the use of the CNN classifier. This classifier makes use of the CNN set of rules for
less complicated identity and separation of rubbish from the organization and for
less complicated separation into respective sections. The dataset is the collection
of daily-­existence wastages like plastic, metal, meals gadgets, etc. Around 450–500
pattern pictures are feed into the dataset for higher capability of set of rules. Each
pattern is zoomed, differed in width and height, mild stage over the object, and a
few greater versions in order that the processing of photo may be made with proper
stage of accuracy.
The convolutional layer works in the following manner – the layer gets a few input
volume, in this example a photo for you to have a particular height, width, and intensity.
There are filters gift which might be essentially matrices which might be initialized
with random numbers at first. The filters are small spatially, however, have intensity
identical as channels of the input photo. For RGB, the filters could have intensity 3; for
grayscale, the tilters could have intensity 1; and so on. The filter is convolved over input
volume. It slides spatially through the photo and computes dot product for the duration
of the photo. The filters come to be generating activation maps for the input photo.

13.6.6.1 Classification
The set of rules works in a two-­segment cycle, the ahead by skip and returned propa-
gation. During the ahead by skip, the photo is handed to each of the above layer, and
the output is calculated. The anticipated output is in comparison with the real output,
and the mistake is calculated. After the mistake is calculated, the set of rules then
adjusts the weights, that is, the spatial values of each of the filters and the biases as
much as the primary enter layer. This adjustment of the weights or the spatial values
of the filters is the returned propagation segment. This returned propagation seg-
ment is used at the side of optimization strategies together with gradient descent to
decrease the mistake as a great deal as possible.

13.6.6.1.1  CNN-based finger-vein extraction

Existing methodologies separate vein patterns by expecting that they create dissemi-
nations, for example, valleys and line portions. In this segment, a CNN is straight-
forwardly prepared to demonstrate the dissemination of vein pixels with practically
no quality dispersion suspicion. To begin with, we name every pixel of a preparation
image as one or the other vein or foundation in view of the joined result of a few
standard check frameworks. For each labeled pixel, a window focused on it is con-
tributed to CNN for preparing. Assuming that a window stretches out external to the
image limit, the missing pixels are incorporated by reflecting. The result of the last
CNN layer is deciphered as the probability of the fix place pixel to have a place with
a vein pattern. Applying CNN to every one of the pixels thusly, the vein patterns are
separated in view of a probability edge. The subsequent binarized images are thus
utilized for finger vein check.
298  Intelligent network design driven by Big Data analytics

Microscopy is a key innovation driving biological disclosure. These days,

microscopy-­based logical discoveries should be validated by quantitative image
analysis
ML is a computational approach to distinguishing patterns in a provided dataset
to make inductions in another, comparable dataset. A traditional textbook model is
the machine recognition of handwriting-­like postal addresses on envelopes. Lately,
conventional article recognition has made colossal advances and is presently mov-
ing toward human precision. Customarily, visual perception by specialists has been
directed to analyze plant illnesses. Nonetheless, there is a risk for blunder because
of abstract insight. In this unique circumstance, different spectroscopic and imag-
ing strategies have been read up for identifying plant infections. In any case, they
require exact instrument and bulky sensors which lead to significant expense and
low effectiveness. All things considered, by and large, customary ML approaches,
for example, SVM and K-­means clustering have complex image preprocessing and
highlight extraction steps, which lessen the efficiency of illness determination.

13.7 Deep learning for lung disease detection

Lung disease detection generally entails categorizing an image as healthy or disease-­

infected. Training is used to create the lung disease classifier, also known as a model.
The process by which a neural network learns to recognize a class of images is
known as training. Deep learning can be used to train a model that can classify
images based on their class labels. As a result, the first step in using deep learning for
lung disease detection is to collect images of lungs with the disease to be classified.
The neural network will then be trained until it can recognize the 14 diseases. The
last step is to categorize new images.
In this case, the model is shown in new images that it has never seen before, and
it predicts the class of those images.
The method for using deep learning to identify lung diseases from medical
images is described. There are three major steps:

1. preprocessing of images
2. education
3. categorization

13.7.1  Preprocessing of images

Image processing is a method of performing operations on an image in order to
improve it or extract useful information from it. It is a type of signal processing in
which the input is an image, and the output can be an image or image characteristics/
features. Image processing is one of today’s fastest growing technologies.
In our proposed system, scanned lung X-­ray images are preprocessed, and the
output is disease prediction. This is accomplished by analyzing the characteristics of
the image that we upload.
 Image Processing for medical images  299

The three basic steps in image processing are as follows:

• importing the dataset image using image acquisition software

• analyzing and manipulating the dataset image
• output, which can be an altered image or a report based on image analysis

13.7.2  Training
Given a training dataset of examples, deep learning neural network models learn to
map inputs to outputs. The training process entails locating a set of weights in the
network that is good, or good enough, at solving the specific problem.
Training data for ML is an important input to algorithms that comprehend
and memorize information from such data for future prediction. However, various
aspects emerge during the ML development process, without which various critical
tasks cannot be completed. Training data is the foundation of any AI or ML project;
without it, it is impossible to train a machine that learns from humans and predicts
for humans.
The process by which a neural network learns to recognize a class of images is
known as training. We must train the neural network until it can recognize diseases.
This is the most important phase of the system because it determines the system’s
accuracy and performance.

13.7.3  Classification
Image classification entails extracting features from an image in order to identify
patterns in a dataset. Using an Artificial neural network (ANN) for image classifi-
cation would be extremely computationally expensive due to the extremely large
trainable parameters.
The images are classified in the final stage. The classification in our system is
of the following types.

1. Normal
2. Pneumonia (19th)
3. Tuberculosis
4. Covid-­19

While uploading the image, our trained model classifies it as one of the diseases
listed above. We use better and optimized scan data in the proposed system to pro-
duce much better and optimized results. Our system improves application perfor-
mance when it comes to computation usage.
It generates hyperfast results without requiring much computation from local,
which saves lives at an earlier stage, and it generates accurate results. The results
produced are highly accurate.
300  Intelligent network design driven by Big Data analytics

13.8 Conclusion

Comparison with alternative approaches in image category, CNNs trounce stand-

ard image process ways in varied applications. This well-­liked fashion is likewise
discovered within the machine-­driven identity of crop illnesses. A number of the
chosen analysis as compared with the performance received with CNNs to it is of
various strategies. Altogether of these research studies, the CNN outcomes are more
than the others. The fine outcomes received for a CNN and for a chance technique in
research that created a comparison.
The distinction of accuracy ranged from 3% to 28.89% using feature fusion
CNN. With the boom of lung illnesses in patients, routinely detecting illnesses and
acquiring correct prognosis through the X-­ray scientific photos turn out to be the
brand new studies recognition within the discipline of PC technological know-­how
and synthetic intelligence to shop the sizable price of guide labeling and classify-
ing. However, the high satisfaction of not unusual place radiograph is’t pleased for
the maximum responsibilities, and conventional strategies are poor to cope with the
large photos.

References

[1] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wire-
less sensor network’. Proceedings of the 5th International Conference on
Communication Systems and Network Technologies, CSNT 2015; Gwalior,
India, IEEE, 2015. pp. 194–200.
[2] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, IEEE, 2014.
[3] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India, IEEE, 2021. pp. 1–6.
[4] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­ based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems; Singapore: Springer; 2020.
[5] Kumar S., Trivedi M.C., Ranjan P., Punhani A. Evolution of Software-­Defined
Networking Foundations for IoT and 5G Mobile Networks. Hershey, PA: IGI
Publisher; 2020. p. 350.
[6] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
 Image Processing for medical images  301

[7] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[8] Kumar V., Arablouei R., Cengiz K., Vimal S., Suresh A. ‘Energy efficient re-
source migration based load balance mechanism for high traffic applications
IoT’. Wireless Personal Communications. 2022, vol. 10(3), pp. 1623–44.
[9] Kumar S., Cengiz K., Trivedi C.M., et al. ‘DEMO enterprise ontology with
a stochastic approach based on partially observable Markov model for data
aggregation and communication in intelligent sensor networks’. Wireless
Personal Communication. 2022.
[10] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering (IJITEE). 2019, vol. 8(6).
[11] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based Sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[12] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring
using machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[13] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[14] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium, Publisher: The Electromagnetics
Academy; 2015. pp. 2363–67.
[15] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
Proceedings of the International Conference on Computational Intelligence
and Communication Networks, CICN 2015; Jabalpur, India, IEEE, 2016. pp.
79–84.
[16] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and iot for smart
road traffic management system’. Proceedings of the IEEE India Council
International Subsections Conference, INDISCON 2020; Visakhapatnam,
India, IEEE, 2020. pp. 289–96.
[17] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. Proceedings of the
Confluence 2020 – 10th International Conference on Cloud Computing, Data
Science and Engineering; Noida, India, IEEE, 2020. pp. 63–76.
[18] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation of
fault tolerance technique for internet of things (iot)’. Proceedings of the 12th
302  Intelligent network design driven by Big Data analytics

International Conference on Computational Intelligence and Communication

Networks, CICN 2020; Bhimtal, India, IEEE, 2020. pp. 154–59.
[19] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 4th International Conference on Information Systems and
Computer Networks, ISCON 2019; Mathura, India, IEEE, 2019. pp. 250–55.
 Chapter 14
IoT-­based architecture for smart health-­
care systems
S P Rajamohana 1, Zbigniew M Wawrzyniak 2, T A Deva
Priyan 3, V Rajeevan 3, N Vaibhav Ram 3, J Vinith 3,
J Keerthanaa 3, and Priyanka Raj 3

Internet of Things (IoT) provides a pathway for connecting physical entities with
digital entities using devices and communication technologies. The rapid growth of
IoT in recent days has made a significant influence in many fields. Healthcare is one
of those fields which will be hugely benefited by IoT. IoT can resolve many chal-
lenges faced by patients and doctors in healthcare. Smart health-­care applications
allow the doctor to monitor the patient’s health state without human intervention.
Sensors collect and send the data from the patient. Recorded data are stored in a
database that enables medical experts to analyze those data. Any abnormal change in
the status of the patient can be notified to the doctor. This chapter aims to study dif-
ferent research works made on IoT-­based health-­care systems that are implemented
using basic development boards. Various hardware parameters of health-­care sys-
tems and sensors used for those parameters are explored. A basic Arduino-­based
health-­care application is proposed using sensors and global system for mobile com-
munication (GSM) module.

14.1   Introduction

In recent years, population growth has caused many problems in the health sector
[1]. On average, around 523  million people suffered through a heart attack in a
year. Doctors cannot provide treatment for them. People who have heart attack has
an average age of greater than 50 years [2]. So, constant monitoring is required for
these kinds of diseases. A smart health-­care application helps in this scenario. The
elder people who cannot travel often can be benefited through this application. It

1
Department of Computer Science & Engineering, Pondicherry University, Coimbatore, India
2
Faculty of Electronics and Information Technology, Warsaw University of Technology, Warsaw, Poland
3
Department of Information Technology, PSG College of Technology, Coimbatore, India
304  Intelligent network design driven by Big Data analytics

Figure 14.1   Block diagram of the proposed system

is also helpful for those children and babies whose both parents have to work [3].
One of the main advantages of this smart health-­care application is that it will be
more helpful for those diseases that take more time to cure and that do not have a
cure [4].
IoT provides a huge benefit in the health sector. It is used to track and monitor
the patients closely from a remote location. Sensors such as heartbeat sensors, pres-
sure sensors, and temperature and humidity sensors are used to record values from
the patients. These values are processed for further analysis to produce better results.
The results that were produced are helpful for further treatment and are stored in a
secured manner [5]. The whole process of storing the values from the sensor can be
done without human intervention.
The connection between wired and intelligent physical devices creates the
major part of the IoT. These devices are capable of performing many functions such
as storing information, collecting information, and also for its processing. These
instruments have further usages such as communication between applications and
the sensors, Internet connectivity, and so on. This smart care health application pro-
vides better accuracy and results than the traditional methods as there are no human
interventions here. It is also proved to be economically better than traditional patient
monitoring [6].
 IoT-­based architecture for smart health care systems  305

Traditional health-­care monitoring requires more human intervention. Humans

make mistakes, and the cost of the traditional method is high. The manpower
involved here is also high.
IoT is used here to address these issues. As the application is automated and
works without any human intervention, the chances of mistakes during this process
are very less. The cost of this IoT-­based patient monitoring is also very less because
of no manpower and the analysis can be done from a remote area [7]. The most
important advantage of IoT is that it will be more helpful in emergency conditions.
The analyzed data will be useful to predict those emergency conditions well before
the seriousness. Thus, smart health-­care applications will be able to reduce the seri-
ousness of those conditions [8]. The Figure 14.1 represents the block diagram of the
proposed system.

14.2   Literature survey

Proposed
Author and framework
publication or prototype Used hardware Used software Communication
S.No. year description components components module Limitations

1. Uday Kumar et al. This paperwork proposes a ✓Arduino ✓Arduino IDE ✓GSM/GPRS Data stored in radio
(2020) [8] smart healthcare patient ✓Blood pressure ✓Embedded C module frequency
data and security system sensor language identification
to monitor patient ✓DS18B20 ✓Thingspeak Server (RFID) tags
health-­care data like Temperature are not secure.
temperature and blood sensor Strong encryption
pressure. The patient strategies should
data are uploaded to the be used to secure
Cloud server. data in the cloud
servers.
2. Vajubunissa Begum This paper aims at helping ✓Arduino UNO ✓Arduino IDE ✓Wi-­Fi module The data collected from
et al. cardiac patients by ✓Temperature sensor ✓Embedded C ✓Bluetooth module this sensor are
(2020) [2] continuously collecting (LM 35) language not secured. The
and monitoring the ✓Humidity sensor ✓Python language cost of this system
health status of the (DHT 11) setup is high.
former using sensors. ✓ECG sensor
This framework also provides (AD8232)
an ECG graph of the ✓Heart rate monitor
patient’s heart rate. sensor (MAX
30105)
✓Body position sensor
(ADXL 335)
✓Raspberry Pi
3. Milon Islam et al. This paperwork is about ✓ESP 32 (node MCU) ✓Thingspeak Server ✓Wi-­Fi module This system is bulkier,
(2020) [9] a smart health-­care module ✓Arduino IDE so it is not flexible
application using ✓Heartbeat sensor to work with.
five sensors. Error ✓Body temperature
percentage in these five sensor (LM35)
sensors is also analyzed ✓Room temperature
in this application. sensor (DHT11)
✓CO Sensor (MQ 9)
✓CO2 Sensor (MQ
135)
4. Dahlia Sam et al. This article proposes a ✓Microcontroller ✓Arduino IDE ✓GSM module The limitation in this
(2020) [3] working IoT-­based (ATMega328P) ✓Cloud database system is the need
architecture that is ✓Arduino UNO for excess wires
capable of monitoring ✓Temperature sensor for connection
the health of any patient (LM35) between the
by using sensors and ✓Blood pressure devices.
microcontrollers during sensor
the pandemic situation.
So, by reducing the
unnecessary expenses
for doctor and hospital
visits.
306  Intelligent network design driven by Big Data analytics

Proposed
Author and framework
publication or prototype Used hardware Used software Communication
S.No. year description components components module Limitations

5. Seena Naik et al. This journal paper gives ✓ECG sensor ✓Cloud database ✓GSM module There are no limitations
(2019) [10] the implementation of ✓Raspberry Pi ✓Raspbian OS mentioned in this
Raspberry Pi and IoT ✓Respiration sensor paper.
in the health system. ✓Acceleration sensor
The various sensors ✓Temperature sensor
gather the body health ✓Blood pressure
parameters information sensor
for the diagnosis ✓Heartbeat sensor
by connecting with
Raspberry Pi, which
is associated with the
cloud and displayed on
the LCD.
6. Suneeta S. Raykar et The proposed framework in ✓Node MCU ✓ThingSpeak Cloud ✓API developed The developed
al. [2019] [11] this paper is ALERT ✓Max30105 oxygen ✓Arduino IDE using MIT App framework allows
(android-­based health saturation sensor inventor the medical experts
enabled ✓ECG sensor AD-­ to examine and
remote terminal). The system 8232 provide advice
is built using node MCU ✓Max30102 pulse rate to one patient
with various sensors for body temperature at a time. This
monitoring the oxygen sensor limitation has to be
level, heart-­pumping changed to multi-­
rate, body temperature user accessibility.
levels.
7. PandiaRajan Jeyaraj The proposed IoT health-­care ✓EEG sensor ✓Cloud for data – The proposed model
et al. (2019) system is combined ✓ECG sensor storage classifies or
[12] with some deep learning Temperature ✓WEKA states the patient
algorithms and sensor sensor Figure 14.1 as
networks. The model ✓Pulse rate sensor more generically
collects the data from ✓NI-­myRIO processor like abnormal,
the patients with the normal, and
help of sensors and subnormal rather
stores them in a cloud than providing
server. Further, the some medical
data are analyzed and assistance.
visualized using the Even though the
DCNN algorithm that proposed system
was developed for provides reliable
learning and using and high accuracy
WEKA analysis is done. health status
prediction, it
does not provide
any portal for the
user (patient) and
medical expert and
communication
platform.
8. Subasish Mohapatra This paperwork proposes ✓Arduino ✓Cloud Server ✓Wi-­Fi module There are no limitations
et al. a health-­care system ✓Heartbeat sensor ✓Embedded C mentioned in this
(2019) [6] using Arduino, collects ✓DS18B20 language paper.
the data from various Temperature ✓Arduino IDE
sensors, and stores sensor
the data into the cloud ✓ ESP8288 (Wi-­Fi
database using a Wi-­Fi module)
module.
9. Abhishek Kumar This paperwork presents a ✓Raspberry Pi ✓Thingspeak server  
✓Wi-­Fi module This system is not
et al. health-­care monitoring ✓DS18B20 ✓Raspbian OS scalable to add
(2018) [7] system to monitor Temperature ✓HTML more sensors, and
temperature and sensor ✓Blynk it also has many
heartbeat using sensors.✓Heartbeat sensor wired connections
This system also has ✓Analog-­to-­digital that make the
a web camera on the converter system less
patient’s side. (MCP3008) flexible.
✓Web camera
10. Shubham Banka et al. This paperwork is about a ✓Raspberry Pi ✓Raspbian OS ✓Wi-­Fi module There are no limitations
(2018) [13] smart health-­care system ✓Temperature sensor ✓GSM module mentioned in this
using Raspberry Pi. (LM35) paper.
This system collects ✓Heartbeat sensor
various details from the ✓Vibration sensor
sensor and is capable of ✓BP sensor
intimating these details
to the patient’s family
and doctor.
 IoT-­based architecture for smart health care systems  307

Proposed
Author and framework
publication or prototype Used hardware Used software Communication
S.No. year description components components module Limitations

11. C. Senthamilarasi, This journal presents a real-­ ✓Arduino UNO ✓Cloud Server ✓Wi-­Fi module There are no limitations
et al. time patient monitoring ✓ECG sensor ✓Arduino IDE mentioned in this
(2018) [14] system interconnected ✓Temperature sensor ✓Embedded C paper.
with IoT to evaluate ✓Heartbeat sensor language
the performance and
practicability of the
systems. This procedure
helps to monitor the
patient’s healthcare
continuously based
on some parameters.
This method will be
supported by Arduino
UNO with a cloud
database.
12. Swaleha Shaikh et al. This paperwork proposes ✓Raspberry Pi ✓Cloud Server ✓Wi-­Fi module Security concerns
(2017) [15] a smart health-­care ✓ LM35 (temp sensor) ✓Raspbian OS over the patient’s
monitoring system ✓Heart rate sensor sensitive data.
using Raspberry Pi and ✓ Blood pressure
a cloud database to sensor
effectively monitor the ✓Accelerometer
patients.
13. Tarannum Khan et al. This paperwork presents ✓Arduino UNO ✓Arduino IDE ✓Wi-­Fi module The data that are
(2017) [16] a patient monitoring ✓Temperature sensor ✓Embedded C uploaded on the
application using ✓Heartbeat sensor Language cloud is insecure.
Arduino UNO. The ✓SD card ✓Blynk Application
stored data are also
presented through the
android application.
14. Shreyaasha This paperwork presents a ✓Arduino UNO ✓Arduino IDE ✓Wi-­Fi module The data stored in the
Chaudhury health-­care application ✓Temperature sensor ✓Embedded C ✓GSM module cloud are insecure,
et al. that monitors health (LM35) language and the cost of
(2017) [5] parameters through ✓ECG sensor ✓HTML this system is also
sensors. An audio ✓Heart rate monitor high.
signaling device and sensor
message service are
attached to the system to
indicate emergencies.
15. Niharika Kumar This paper proposes a smart ✓Arduino UNO ✓Arduino IDE ✓Wi-­Fi module The cost of this
(2017) [17] frame for a healthcare ✓Gyroscope ✓Embedded C proposed system
system that uses ✓ECG sensor Language is high.
architecture protocols of ✓Temperature sensor ✓HTML
the 6LoWPAN protocol ✓ Heart rate sensor
and IEEE 11073, which
is associated with cloud
networks.
16. Fatima Alshehri et al. This paper provides a – – – –
[2020] [18] detailed survey about
smart healthcare. This
paper includes various
methods for smart
healthcare like the IoT,
Internet of medical
things, and artificial
intelligence.
17. Yasmeen Shaikh et al. This paperwork proposes – – –  –
(2018) [19] several methods for
providing health-­care
solutions using IoT.
The major methods
discussed in this paper
are RFID and named
data networking.
18. Wei Li et al. This paper is about a survey – – – –
(2020) [20] that explores the
different possibilities
of implementing
smart healthcare using
machine-­learning and
big data analysis.
19. Ramakrishna Hegde The proposed system is – – – –
et al. to monitor the daily
(2021) [21] health-­related activities
of a patient and report to
the doctors through the
internet with a different
approach.
308  Intelligent network design driven by Big Data analytics

Proposed
Author and framework
publication or prototype Used hardware Used software Communication
S.No. year description components components module Limitations

20. Hongxu Zhu et al. This paper framework is – – – –

(2019) [22] about the various
phases in healthcare
with IoT along with the
management of patient
health data and having
an oriented management
system.
21 Durga Amarnath M. This paperwork proposes  
✓Microcontroller  
✓SQL ✓Wi-­Fi module The proposed system
Budida et al. an end to end well-­ ATMEL 89s52  
✓HTML is outdated and
(2017) [23] organized health  
✓Temperature requires regular
application. The sensor maintenance.
proposed framework  
✓Pulse rate sensor
uses ATMEL  
✓RS-­232
microcontroller as the  
✓Analog-­to-­digital
main processor. It uses converter
HTML for organizing  
✓Voltage regulator
the web pages and IC-­7805
uses SQL for storing
the data.
22 Kavita Jaiswal et al. This paper is about a  
✓Raspberry Pi ✓SQL ✓Wi-­Fi module The proposed system is
(2017) [24] health-­care monitoring  
✓Temperature ✓Docker costly since it has
system which uses sensor a server that has a
Raspberry Pi as a main  
✓ECG docker container
microcontroller. It uses  
✓Blood and a database.
a Docker container and  
✓Pressure sensor
a DBMS to manage the
data created.
23  Gowrishankar, S. This paper uses sensors  
✓Pulse sensor  ✓ESP8266  
✓Wi-­Fi  Data transfer between
M.Y. Prachita et to collect the state of  
✓LM 35  module devices must be
al. (2017) [25] the heart and body  
✓Temperature fast enough with
temperature with sensor minimized delay
different test cases  
✓Arduino UNO response, sensor
and provides an alert module should
message using an LCD extract accurate
monitor. readings, and two
different sensors
must be used and
should be under
little control of
microcontroller
to be accurate in
output.
24 Mayur, Suraj, The paper proposes a  
✓ECG monitor,  
✓NA ✓Mobile data/Wi-­  Wi-­Fi module is used
 
Shubham, Nikhil portable ECG monitor Android phone Fi module to transfer and
et al. (2016) where the heart rate will monitor data, so
[26] be collected and sent to connectivity must
an android phone. The be constant and
Android phone is using very fast enough to
a detection algorithms to avoid delay. Sensor
find out the abnormality. module should
extract accurate
readings. Data
are stored in the
server.
25 Manisha, Mamidi This paper proposes a  
✓MI Band 2,  
✓NA ✓Mobile data/Wi-­  The response time is
 
(2016) [27] method to record the android phone, big Fi module limited, when the
patient’s heart and pulse data analytics micro-­controller
rate which is further transmits
analyzed using big data the information
analytics. the encoder
must respond
immediately and
it needs precise
decoding.
 IoT-­based architecture for smart health care systems  309

The detailed literature survey analysis is given in Table 14.1.

14.3   Related works

Uday Kumar et al. [8] proposed Arduino-­based patient monitoring system that stores
patient details such as blood pressure and temperature. The author implemented this
system by using a temperature sensor to monitor temperature and a blood pressure
sensor to monitor blood pressure. Patient data that are collected through sensors are
stored in radio frequency identification (RFID) tags and are also uploaded in the
cloud. The stored data will be useful to doctors to analyze patients’ conditions. This
system has GSM/General Packet Radio Service (GPRS) module to alert emergen-
cies to the patient’s side by short message services (SMS).
Vajubunnisa Begum et al. [2] proposed a health-­care monitoring system spe-
cially designed for cardiac patients using Raspberry Pi. The data are collected
through the sensors such as temperature sensors, heartbeat sensors, ECG sensors,
and body position sensors. Wi-­Fi module is used for sending the collected data from
sensors to the cloud. The data are also transferred to the doctor’s side using the
Bluetooth module. Serial plotter software is used to plot the stored data collected
from the ECG sensor. The analyzed result from the sensor is displayed on the LCD.
Milon Islam et al. [9] proposed IoT-­based health-­care application using Node
MCU (which is also known as ESP 32). Five sensors collected patient’s data and
there are a heartbeat sensor to monitor heartbeat, a room-­temperature sensor to mon-
itor room temperature, a body temperature sensor to monitor temperature, an MQ9
sensor to monitor CO gas, and an MQ135 sensor to monitor CO2 gas.
Dahlia Sam et al. [3] proposed an Arduino-­based model patient monitoring sys-
tem to monitor the patient’s situation. Patient monitoring is done using a tempera-
ture sensor for monitoring body temperature and an optical sensor that measures
heartbeat in pulsates. These sensors send the digital signals to the Arduino UNO,
and sensor data through a Wi-­Fi module and can be accessed through the cloud by
doctors and also relatives.
Seena Naik et al. [10] have proposed a framework by implementing a working
model that monitors and checks a patient’s health condition even from a distance in
a cheap way using an ECG sensor, temperature sensor, and heartbeat sensor condi-
tioned by Raspberry Pi. All the digital signals will be transmitted to the Raspberry Pi
and stored in a database accessed by the cloud. The details will be sent to the health
specialist via the Wi-­Fi module.
Suneeta S. Raykar et al. [11] have proposed a duplex communication system
that provides a gateway to monitor various health parameters. The data will be
recorded using a sensor and sent to the ThingSpeak cloud. The medical experts will
use the data to make decisions on the patient’s health. They can also give medical
recommendations based on that data. The recommendations that are given by the
medical experts can be seen by the patient through an android application that is
310  Intelligent network design driven by Big Data analytics

developed using MIT App inventor. It also suggests future modifications that can be
made for the proposed framework.
PandiaRajan Jeyaraj et al. [12] have proposed a patient monitoring system along
with a deep learning algorithm for making an accurate prediction based on the patient
collected data. The system collects various data such as EEG, ECG, body tempera-
ture, and vital signs. The data are connected using an intelligent sensor network,
and then the data are stored using central cloud storage with the help of the myRIO
processor along with a Wi-­Fi module. The data from the cloud are taken, classified
using WEKA, and analyzed using a Deep Convolutional Neural Network (DCNN)
learning algorithm. The model provides results with 97.2% prediction accuracy.
Swaleha Shaikh et al. [6] proposed an application that observes the patient’s
health conditions with Raspberry Pi at its heart. This framework presents a system
in which the data are collected using embedded sensors that are wearable, and the
health status of the patient is monitored concerning some parameters dynamically.
The collected data are then transmitted to the Raspberry Pi, which will process and
analyze the data. These data that have been analyzed by the processor are stored in
the cloud. The stored results are then used by the doctors when the details of the
patient are needed.
Abhishek Kumar et al. [7] proposed a smart health-­care application using
Raspberry Pi. Raspberry Pi is connected with a temperature sensor to monitor tem-
perature and heartbeat sensors to monitor the heartbeat. The input from the sensor
is processed in Raspberry Pi and displayed on LCD. The data are then sent to the
cloud by using Raspberry Pi that enables doctors to continuously monitor the status
of the patient. The patient’s side has a web camera that helps the doctor to monitor
the patient. Both Web and android application is used to present the data collected
through the sensor.
Shubham Banka et al. [13] proposed a smart health-­care monitoring system
using Raspberry Pi. The author implemented this system using a temperature sensor,
heartbeat sensor, BP sensor, and vibration sensor to monitor the patient’s tempera-
ture, heartbeat, BP, and shivering of the patient. The data collected from the sen-
sor are stored and are presented through the web user interface. A GSM module is
attached to inform the critical situations of the patients through SMS.
Senthamilarasi et al. [14] proposed a model for monitoring patients’ healthcare
by Arduino UNO. The real-­time implementation of health-­care applications uses a
temperature sensor for body temperature, ECG for heartbeat, and a heartbeat sensor
for heartbeat rate monitoring. These sensors send the signal to the Arduino UNO,
and sensor data can be monitored by any smart device that is interconnected with the
cloud database, which acts as a server for communication. The paper also discusses
the recent advancements in the IoT-­based health-­care eco-­system.
Subasish Mohapatra et al. [15] implemented a health-­care monitoring system
using Arduino. This model uses a temperature sensor and pulse sensor that are con-
nected to the Arduino. The health parameters given by the sensors are then stored in
the cloud database by using Wi-­Fi module. This cloud database processes the data
provided by the sensors and if the data exceeded the threshold value, it makes an
emergency call to the concerned doctor with the current health status of the patient
 IoT-­based architecture for smart health care systems  311

and with a full detailed medical report, so that the doctor suggests a proper health-­
care measures to be taken in case of critical health conditions.
Tarannum Khan et al. [16] proposed an Arduino-­based patient monitoring sys-
tem to monitor the health of the patients. The patient’s details are collected using
two sensors that are temperature sensor and heartbeat sensor to monitor temperature
and heartbeat. An LCD screen is used to display the monitored details. The collected
details are uploaded on the server and are converted into JSON links for visualizing
in an android application.
Shreyaasha Chaudhury et al. [5] proposed a patient monitoring system using
the IoT. The author implemented this system using Arduino UNO, temperature sen-
sor, ECG sensor, and heartbeat sensor to monitor the patient’s health parameter.
The data collected through the sensors are monitored and analyzed by the doctor.
Abnormal changes in the health parameters are notified to the doctor using a buzzer
and SMS. The data collected are stored in a database and are present through a
simple web page.
Niharika Kumar [17] gave a framework end-­to-­end health-­care system associ-
ated with Arduino UNO. The implementation of the health-­care application system
uses a temperature sensor for body temperature, ECG for heartbeat, gyroscope for
sense angular velocity, and heartbeat sensor for the heartbeat rate monitoring. These
sensors send the signal to the Arduino UNO, which is connected to the Internet and
acts as the communication media between the IoT devices and exhibit devices such
as a computer system or phones.
Durga Amarnath M. Budida et al. [23] proposed a health-­care system that uses
ATMEL 89s52 microprocessor as its base. The system contains a login to validate
the user who can access the system. The data are sent from microprocessor to other
computers in the hospital side using Wi-­Fi module. The system has a temperature
sensor and a pulse rate sensor to monitor the patient’s health, and it also requires
analog to digital convertor. This convertor will convert the analog to digital on
which the microprocessor works.
Kavita Jaiswal et al. [24] implemented a Raspberry Pi-­based health monitoring
system that is used to monitor the health condition of a patient from a remote loca-
tion. This system uses a Docker container to manage the data that are collected and
also uses a Database to store the data about the patient. Once the data is stored, it can
be accessed by multiple users who can be benefitted.
Gowrishankar et al. [25] proposed and designed a framework that determines
the human body temperature and heartbeat of the patient, then it redirects the data to
the server end by using a cost-­efficient microcontroller with great effect. It uses three
different sensors, such as pulse sensors, heartbeat sensor, and temperature sensor,
and they are controlled by the microcontroller.
312  Intelligent network design driven by Big Data analytics

14.4   Hardware components and sensors

14.4.1   Development boards

Development boards are circuit boards used to test sensors and act as an interface
between the sensors and the processing units. The development boards used in the
above research paper works are Arduino UNO, Raspberry Pi, NI-­my RIO processor,
and Node MCU ESP8266.

14.4.1.1   Arduino UNO

Arduino UNO is an open-­source microcontroller development board used widely in
IoT projects. The major benefits of using Arduino are low cost and flexibility. The
Arduino UNO is used to monitor the health parameters in healthcare applications
proposed in References 2, 3, 8. In Reference 2, Arduino is used for developing a
health-­care application for an electrocardiogram to detect the electrical signals of
our heart.

14.4.1.2   Raspberry Pi
Raspberry Pi is a single-­board computer that allows us to connect and explore with
different sensors. In References 6, 7, 10, the health-­care application is built upon the
Raspberry Pi platform. In Reference 2, the smart health-­care application is imple-
mented in both the Arduino and Raspberry Pi platforms to compare the efficiency.
Raspberry Pi gave better results than the Arduino in the patient monitoring system
proposed in Reference 2.

14.4.1.3   NI-my RIO processor

NI-­my RIO processor is an evaluation board that is used to create an application
using the microprocessor present in it. In Reference 12, patient monitoring is done
with the help of NI-­my RIO processor. The data collected here are stored and ana-
lyzed using machine learning algorithms.

14.4.1.4   Node MCU ESP 8266 module

Node MCU is an open-­source development board and is extensively known for its
Wi-­Fi chip present in it. Node MCU is used as a development board in the health-­
care monitoring system proposed in References 9 and 11.

14.4.2   Sensors
14.4.2.1   Blood pressure sensor
The blood pressure sensor is capable of monitoring the blood flow of a patient. The
pressures created in the blood flow are detected and converted to electrical signals.
Blood pressure is an important factor when considering a person’s health. This sen-
sor is used in the patient monitoring system proposed in References 7, 8, 10 to study
the blood pressure of the patient.
 IoT-­based architecture for smart health care systems  313

14.4.2.2   Temperature sensor (DS18B20)

The temperature of the place and patient is one of the considering factors that can
affect a patient’s health. This temperature sensor is digital, programmable, and also
follows a single wire protocol. The temperature of the patient is collected using
this sensor in frameworks proposed in References 7–9 to monitor the patient’s
temperature.

14.4.2.3   Temperature sensor (LM35)

Temperature sensor (LM35) works on the principle of the diode. The output is in the
form of voltages. This temperature sensor helps take into account the temperature
required for the patient’s health. This temperature sensor is used in the health-­care
application proposed in References 3, 9, 12.

14.4.2.4   Humidity sensor (DHT11)

Humidity sensor acts as a digital temperature sensor as well as a humidity sensor. It gives
a digital signal as output by measuring the air humidity. This sensor is used to study the
room temperature in the patient monitoring system proposed in Reference 9.

14.4.2.5   ECG sensor (AD-8232)

ECG sensor is used to monitor heartbeats. The condition and activity of heart pump-
ing and vibrations are measured and electrically converted to a signal that can be
monitored on a display. In References 2, 11, 12, the activity of the heart is monitored
using an ECG sensor.

14.4.2.6   Heart rate monitor sensor (MAX 30105)

Heart rate monitor sensor (MAX 30105) is used for heartbeat plotting and heart
rate monitoring. It has three LEDs for pulses and detects the reflection accordingly.
Heart rate is monitored using this sensor in frameworks proposed in References 2,
9, 10 to monitor heart rate.

14.4.2.7   Body position sensor (ADXL 335)

The body position sensor is also called an accelerometer. The output of this sensor
is analog voltages that are proportional to the acceleration. By measuring the accel-
eration, it can identify the position or angle of the vibration. This sensor is used in
health-­care applications proposed in frameworks [2, 10, 15] to determine the body
position of the patient.

14.4.2.8   CO sensor (MQ 9)

CO sensor is used for gas leakage detection, such as LPG, CO, CH4. Its sensitivity
can be adjusted by using a potentiometer. It has a fast response time and high sensi-
tivity. In Reference 9, this sensor is used to monitor CO gas.
314  Intelligent network design driven by Big Data analytics

14.4.2.9   CO2 sensor (MQ 135)

In CO2 sensor (MQ 135), the sensitivity can be set by using a potentiometer. It can
read gases such as ammonia nitrogen, oxygen, and also alcohol. Not like MQ9, it is
capable of detecting more harmful gases. This sensor is used in the patient monitor-
ing system proposed in Reference 9.

14.4.2.10 Respiration sensor
A respiration sensor is used for giving the output of respiration vibration in the form
of a wave. As the breathing of the patient takes place, a respiratory waveform can
be shown in the display. In Reference 10, this sensor is used to monitor the patient’s
respiration.

14.4.2.11   Pulse rate body temperature sensor (MAX 30102)

Pulse rate body temperature sensor is a heart rate monitor biosensor and also a pulse
oximetry sensor. It measures the heart rate of the patient and also the hemoglobin
level of a patient by measuring oxygen. This sensor is used in frameworks proposed
in References 11, 12 to monitor the pulse rate.

14.4.2.12   Vibration sensor

The vibration sensor works in the same way as the pulse sensor, but it measures the
vibration in terms of frequency. This is used to detect any breakdowns or imbalances
present in the patient. This sensor is used in the patient monitoring system proposed
in Reference 13 to monitor the shivering of patients.

14.4.2.13   Gyroscope sensor

Shortly known as a gyro sensor, gyroscope can measure angular velocity. This sen-
sor can be used for maintaining the stability of a patient and measuring the level of
effectiveness. In Reference 17, this sensor is used to maintain the stability of the
patient.

14.4.3   Other modules

14.4.3.1   Radio frequency identification
RFID is a technology that is used to automatically identify and collect information
about a particular object. RFID uses radio waves to identify and collect data about
the object. In Reference 8, RFID technology is used to store a particular patient’s
details in RFID tags.

14.4.3.2   GSM/GPRS module

GSM/GPRS module is the communication module that is used to establish a connec-
tion with mobile mostly in Arduino-­based frameworks. The GSM module is capable
of sending SMS, and the GPRS module is used for providing network connectivity.
 IoT-­based architecture for smart health care systems  315

This module is used in health-­care systems proposed in References 3, 8, 10 to notify

the doctors during emergencies.

14.5   Proposed work

The proposed health-­care application uses Arduino at its heart which is a microcon-
troller capable of articulating multiple sensors. Arduino development boards have
the power of transforming the data provided to them. The system uses three sensors
to monitor the health of the patients. The sensors are a temperature sensor, heartbeat
sensor, and pressure sensor [28]. Many varieties of temperature sensors are used to
read the temperature from patients. The temperature sensor used in this proposed
framework is LM 35. A heartbeat sensor is used for monitoring the heartbeat of the
patient. The vibration sensor is capable of checking the shivering of the patient.
The data collected from Arduino are constantly monitored. The GSM module sends
messages to doctors when the patient’s state is critical. The LCD is used to display
the temperature, heartbeat, and vibration status of a patient. The data can also be
accessed online with the help of a Wi-­Fi module [27, 29–43].

Figure 14.2   Workflow of the proposed health-­care application

316  Intelligent network design driven by Big Data analytics

Figure 14.3   Arduino UNO

The workflow of the proposed system is shown in Figure 14.2. The objective

of this work is to monitor the patient’s health parameters continuously. This system
uses an Arduino microcontroller for interfacing different sensors [44–47]. First, the
sensor modules are initialized. The code should be written in Arduino IDE using
embedded C language for these sensor modules. The code is compiled and verified
to check the errors. In case of any errors, the code is modified appropriately. Once
the code is error-­free, it is uploaded to the Arduino development board. The health
parameters such as temperature, pressure, and heartbeat are red using sensors. The
data collected from the patient are now processed. If there exist any abnormalities in
a patient’s health, messages are sent to the doctor with the help of the GSM module
[3, 8, 10].

14.5.1   Hardware components used

14.5.1.1   Arduino
The microcontroller used for the project is Arduino UNO. It has 6 analog pins and
14 digital pins. It is cheaper and has many user-­friendly features. It has simple archi-
tecture and supports higher input/output drive strength (Figure 14.3).

14.5.1.2   Temperature sensor

The temperature sensor used in this project is LM-­35. It has three terminals, namely
VCC, GND, and OUT. It can measure temperature ranging between –55 and 150°C.
It measures analog voltage that is proportional to the temperature. We can convert
 IoT-­based architecture for smart health care systems  317

Figure 14.4  Temperature sensor

the analog voltage to digital format for readability purposes. The temperature sensor
is shown in Figure 14.4.

14.5.1.3   Heartbeat sensor

A heartbeat sensor is used to measure the heartbeat/pulse of a human. It is also
known as a pulse sensor and can detect the heart rate of a person 3 feet away. The
sensor detects the heartbeat with the help of passing green light on the finger and
the methodology is known as photoplethysmogram. The patient’s or user’s finger-
print is kept on the front side of the sensor (heart-­shaped side) and an LED to indi-
cate the functioning of the sensor [48]. On the backside, we have resistors, diodes,
microchips, and capacitors. The sensor can handle 3.3–5 V and a current of 4 mA.
It has three male header connectors—one for VCC, another for GND, and another
for signal output—that is connected to the analog input of the Arduino board. The
heartbeat sensor is depicted in Figure 14.5.

14.5.1.4   Vibration sensor

We are using vibration sensor module SW-­420 to check the shivering level of the
patient. This sensor can detect the vibration in the human body beyond a specified
limit/threshold. It has three pins Vcc, GND, and a digital output pin. It is com-
bined with an LM393 comparator to detect the vibration. A potentiometer is pre-
sent to adjust the threshold of vibration to detect. The vibration sensor is shown in
Figure 14.6.

14.5.1.5   GSM/GPRS module

GSM/GPRS module is responsible for the communication between the development
board and the network. GSM module allows the users to send messages by using
SMS. GPRS module is an advancement in the GSM module that allows the user to
connect to the Internet. The GSM module is depicted in Figure 14.7.
318  Intelligent network design driven by Big Data analytics

Figure 14.5   Heartbeat sensor

14.5.1.6   LCD display

The LCD of the Arduino is used to display the content from the Arduino board. It is
a 16 × 2 LCD module that has four connection ports. These ports are responsible for
connection purposes. This display is capable of displaying 16 characters in 2 lines.
The LCD is shown in Figure 14.8.

Figure 14.6   Vibration sensor

 IoT-­based architecture for smart health care systems  319

Figure 14.7   GSM/GPRS module

14.5.1.7   Wi-Fi module (ESP8266)

ESP8266 module helps the microcontroller to connect with Wi-­Fi network. It is an
SOC (System On-­chip) integrated with a TCP/IP protocol stack, which can provide
microcontroller access to any type of Wi-­Fi network. This can also host an applica-
tion individually. ESP8266 module generally has 16 GPIO pins. It also allows us
to enable connections such as SPI, I2C, UART, and PWM pins. Figure 14.9 shows
Wi-­Fi module [49].

Figure 14.8   LCD display

320  Intelligent network design driven by Big Data analytics

Figure 14.9   ESP8266 module

14.6   Implementation and results

The proposed system is implemented using the Proteus application. The different
components used in this application are connected to the Arduino. The block dia-
gram of the proposed system is demonstrated in Figure 14.2. The sensor modules
used in the system are connected to the Arduino development board using wires.
Each sensor used in this system contains a Vcc and a ground. These connections are
provided to the sensors. The output pin of the LM35 sensor is connected to A0 of
the Arduino board. The output pin of the heartbeat sensor is connected to A2 of the
Arduino board. D0 of the Arduino board is connected to the output pin of the vibra-
tion sensor. Virtual terminal in Proteus is used to display the output of the proposed
system. The RX pin of the serial monitor is connected to D1 of Arduino. The GSM
module contains two pins, RX and PX. The RX pin and TX pin of the GSM module
is connected to D2 and D3 of the Arduino board, respectively. The circuit diagram
of the proposed system is shown in Figure 14.10.
The code for the proposed system is compiled and executed in the Arduino
IDE. The hex file of the executed code is uploaded to the Arduino board that is the
 IoT-­based architecture for smart health care systems  321

Figure 14.10   The circuit diagram of an implemented system in Proteus

program file of the board. The program file of the GSM module is uploaded in it. The
proposed system is simulated to view the results.
The data collected from the sensors are displayed on the virtual terminal. First,
the temperature is displayed in two units, and they are Celsius and Fahrenheit. The
output from the vibration sensor is either the patient is normal or the patient is shiv-
ering. The heartbeat sensor gives heartbeats per minute. If the patient is normal,
then there will be no alert messages sent to the doctor’s side. The data displayed are
shown in Figure 14.11.
Each parameter is set to a threshold value. When the data recorded from the sen-
sor exceed the threshold value, alert messages are sent to the doctor. The threshold

Figure 14.11   Data displayed in virtual terminal

322  Intelligent network design driven by Big Data analytics

Figure 14.12   Alert message sent to doctor’s side

value for the temperature sensor is 101 °F. If the temperature value exceeds the
given value, then the alert messages are sent to the doctor using the GSM module.
In the same way, the threshold value of the heartbeat sensor is 50. If the heartbeats
per minute reduce below 50, then the alert messages should be sent to the doctor’s
number. The alert message is shown in Figure 14.12.
Figure 14.13 shows the temperature reading of the patient collected during dif-
ferent times of the day. For this particular patient, the temperature value is between
98.2 and 98.9 °F.
The heartbeat of the patient recorded using sensors is shown in Figure 14.14. .
The heartbeats of the patient vary from range 69 to 72.

Figure 14.13   Graphical representation of temperature data

 IoT-­based architecture for smart health care systems  323

Figure 14.14   Graphical representation of heartbeat data

14.7   Conclusion
IoT combined with the health-­care sector has proved to make remote monitoring
of patients possible and effective. It unleashes the power and possibility to keep
the patients healthy and also empower the medical experts to deliver instant care to
the patients. It also prevents unnecessary admission in hospitals since every single
change in their body is being collected at an early stage and remedies can be pro-
vided superlatively.
IoT in the health-­care system is proving to be significantly effective and the
outcome is very much beneficial, not only to the patients but also to the medical
experts and hospitals. This paper deals with various health-­care monitoring systems
that have been developed using IoT and provides an extensive survey on the same.
It briefs about different designs for the health-­care system. Also, the different types
of IoT components used for building the system, their limitations, and support pro-
vided by each type of design have been extensively discussed. An Arduino-­based
application is simulated on the Proteus application.

References

[1] Zhu H., Wu C.K., Koo C.H., et al. ‘Smart healthcare in the era of Internet-­of-­
Things’. IEEE Consumer Electronics Magazine. 2019, vol. 8(5), pp. 26–30.
[2] Vajubunnisa Begum R., Dharmarajan K. ‘Smart healthcare monitoring sys-
tem in IoT’. European Journal of Molecular & Clinical Medicine. 2020, vol.
7(4).
324  Intelligent network design driven by Big Data analytics

[3] Dahlia Sam S., Srinidhi V., Niveditha R., Amutha S. ‘Progressed IOT based
remote health monitoring system’. International Journal of Control and
Automation. 2020, vol. 13(2s).
[4] Baker S.B., Xiang W., Atkinson I. ‘Internet of things for smart healthcare:
technologies, challenges, and opportunities’. IEEE Access. 2017, vol. 5, pp.
26521–44.
[5] Chaudhury S., Paul D., Mukherjee R., Haldar S. ‘Internet of thing based health-
care monitoring system’. 2017 8th IEEE Annual Information Technology,
Electronics and Mobile Communication Conference (IEMCON), Publisher:
IEEE, Vancouver, BC, Canada; IEEE, 2017.
[6] Mohapatra S., Mohanty S., Mohanty S. Smart healthcare: an approach for
ubiquitous healthcare management using iiott, Editor(s): Nilanjan Dey,
Himansu Das, Bighnaraj Naik, Himansu Sekhar Behera, in advances in
ubiquitous sensing applications for healthcare, big data analytics for in-
telligent healthcare management, academic press, 2019, 175-­196, ISBN
9780128181461. Elsevier; 2019.
[7] Kumar A., Chattree G., Periyasamy S. Smart healthcare monitoring system,
wireless personal communications, 453–463 (2018). https://​doi.​org/​10.​1007/​
s11277-­​018-­​5699-­​0. Vol. 101. Springer; 2018. pp. 453–63.
[8] Uday Kumar K., Shabbiah S., Rudra Kumar M. ‘Design of high-­security
smart health care monitoring system using IoT’. International Journal of
Emerging Trends in Engineering Research. 2020, vol. 8(6).
[9] Islam M.M., Rahaman A., Islam M.R. ‘Development of smart healthcare
monitoring system in IoT environment’. SN Computer Science. 2020, vol.
1(3), p. 185.
[10] Seena Naik K., Sudarshan E. ‘Smart healthcare monitoring system using
raspberry Pi on IoT platform’. ARPN Journal of Engineering and Applied
Sciences. 2019, vol. 14(4).
[11] Raykar S.S., Suneeta S., Shet V.N. ‘Design of healthcare system using iot
enabled application’. Advanced Materials for Clean Energy and Health
Applications, Today: Proceedings, Volume 23, Part 1, 2020, Pages 62-­67.
2019.
[12] Rajan Jeyaraj P., Nadar E.R.S., Jeyaraj P. ‘Smart-­monitor: patient monitoring
system for IoT-­based healthcare system using deep learning’. IETE Journal
of Research. 2019, vol. 16(1), pp. 1–8.
[13] Banka S., Madan I., Saranya S.S. ‘Smart healthcare monitoring using IoT’.
International Journal of Applied Engineering Research. 2018, vol. 13(15).
[14] Senthamilarasi C., Jansi Rani J., Vidhya B., Atitha H. ‘A smart patient health
monitoring system using IoT’. International Journal of Pure and Applied
Mathematics, vol. 119(16).
[15] Shaikh S., Chitre V. ‘Healthcare monitoring system using iot’. International
Conference on Trends in Electronics and Informatics ICEI, Publisher:
IEEE,Tirunelveli, India, DOI: 10.1109/ICOEI.2017.8300952; 2017.
[16] Khan T., Chattopadhyay M.K. ‘Smart healthcare monitoring system’. IEEE,
International Conference on Information, Communication, Instrumentation
 IoT-­based architecture for smart health care systems  325

and Control (ICICIC), Publisher: IEEE, Indore, India, DOI: 10.1109/

ICOMICON.2017.8279142; 2017.
[17] Kumar N. ‘IoT architecture and system design for healthcare sys-
tems’. International Conference on Smart Technology for Smart
Nation IEEE, Publisher:IEEE, Bengaluru, India, DOI: 10.1109/
SmartTechCon.2017.8358543; 2017.
[18] Alshehri F., Muhammad G. ‘Comprehensive survey of the IoT and AI-­Based
smart healthcare’. IEEE Access: Practical Innovations, Open Solutions. 2021,
vol. 9, pp. 3660–78.
[19] Shah Y., Parvati V.K., Biradar S.R. ‘Survey of smart healthcare systems using
internet of things (iot)’. 2018 International Conference on Communication,
Computing and Internet of Things (IC3IoT), Publisher: IEEE, Chennai,
India, DOI:10.1109/IC3IoT.2018.8668128; IEEE, 2018.
[20] Li W., Chai Y., Khan F., et al. ‘A comprehensive survey on machine learning-­
based big data analytics for IoT-­enabled smart healthcare system’. Mobile
Networks and Applications. 2021, vol. 26(1), pp. 234–52.
[21] Hegde R., Ranjana S., Divya C.D. ‘Survey on development of smart health-
care monitoring system in iot environment’. Fifth International Conference
on Computing Methodologies and Communication, Publisher: IEEE, Erode,
India, DOI: 10.1109/ICCMC51019.2021.9418405; ICCMC, 2021.
[22] Zhu H., Wu C.K., Koo C.H., et al. ‘Smart healthcare in the era of Internet-­of-­
Things’. IEEE Consumer Electronics Magazine. 2019, vol. 8(5), pp. 26–30.
[23] Budida D.A.M., Mangrulkar S. D.R. ‘Design and implementation of smart
healthcare system using iot’. International Conference on Innovations in
Information; Embedded and Communication Systems (ICIIECS), Publisher:
IEEE, Coimbatore, India, DOI: 10.1109/ICIIECS.2017.8275903, 2017.
[24] Jaiswal K., Sobhanayak S., Mohanta B.K., Jena D. ‘IoT-­ cloud based
framework for patient’s data collection in smart healthcare system using
raspberry-­pi’. International Conference on Innovations in Information,
Embedded and Communication Systems (ICIIECS), Publisher: IEEE, Ras
Al Khaimah, United Arab Emirates, DOI: 10.1109/ICECTA.2017.8251967;
2017.
[25] Gowrishankar S.M.Y., Prakash P.A. 'IoT based heart attack detection, heart rate
and temperature monitor'. International Journal of Computer Applications,
26-­30, DOI: 10.5120/Ijca2017914840, July 2017. 2017, vol. 170(5).
[26] Arith B., Deepak K.C., Sathish K., Aboobacker S., Kumar A. ‘Heartbeat sens-
ing and heart attack detection using internet of things: IoT’. International
Journal of Engineering Science and Computing. 2017, vol. 7(14), pp. 6662–6.
[27] Manisha M., Neeraja K., Sindhura V., Ramya P. ‘IoT on heart attack detection
and heart rate monitoring’. International Journal of Innovation in Engineering
and Technology (IJIET). 2020.
[28] Jaiswal K., Sobhanayak S., Turuk A.K., Bibhudatta S.L., Mohanta B.L.,
Jena D. ‘An IoT-­ cloud based smart healthcare monitoring system using
the container-­ based virtual environment in Edge device’. International
Conference on Emerging Trends and Innovations in Engineering and
326  Intelligent network design driven by Big Data analytics

Technological Research (ICETIETR); Ernakulam, India, 11–13 July 2018;

2018. pp. 1–7.
[29] Kumar S., Ranjan P., Ramaswami R. ‘EMEEDP: enhanced multi-­hop energy
efficient distributed protocol for heterogeneous wireless sensor network’.
Proceedings – 2015 5th International Conference on Communication Systems
and Network Technologies CSNT; Gwalior, India, 04–06 April 2015; 2015.
pp. 194–200
[30] Kumar S., Rao A.L.N., Ramaswami R. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, 07–08 February 2014; 2014. pp. 350–5.
[31] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, Internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); 2021. pp. 01–6.

[32] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­ based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems; Springer, Singapore; 2019.
[33] Kumar S., Trivedi M.C., Ranjan P. Evolution of software-­defined network-
ing foundations for iot and 5G mobile networks,publisher: IGI, DOI:
10.4018/978-­1-­7998-­4685-­7, ISBN13: 9781799846857; 2020. p. 350.
[34] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[35] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[36] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications IoT’. Wireless
Personal Communications. 2021, vol. 10(3).
[37] Kumar S., Cengiz K., Trivedi C.M., et al. ‘DEMO enterprise ontology with
a stochastic approach based on partially observable Markov model for data
aggregation and communication in intelligent sensor networks’. Wireless
Personal Communication. 2022.
[38] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic to-
rus network-­on-­chip architecture’. International Journal of Innovative
Technology and Exploring Engineering (IJITEE). 2019, vol. 8(6), pp.
2278–3075.
[39] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
 IoT-­based architecture for smart health care systems  327

[40] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring
using machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[41] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[42] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium. 2015, vol. 2015, pp. 2363–7.
[43] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
Proceedings – 2015 International Conference on Computational Intelligence
and Communication Networks, CICN 2015; 2016. pp. 79–84.
[44] Sharma A., Awasthi Y., Kumar S. ‘The role of Blockchain, AI and IoT for
smart road traffic management system’. Proceedings – 2020 IEEE India
Council International Subsections Conference, INDISCON 2020; 2020. pp.
289–96.
[45] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. Proceedings of the
Confluence 2020 – 10th International Conference on Cloud Computing, Data
Science and Engineering, Publisher: IEEE, Noida, India, DOI: 10.1109/
Confluence47617.2020.9057829; 2020. pp. 63–76.
[46] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation of
fault tolerance technique for Internet of things (IoT)’. Proceedings – 2020 12th
International Conference on Computational Intelligence and Communication
Networks, CICN 2020; 2020. pp. 154–9.
[47] Reghu S., Kumar S. ‘Development of robust infrastructure in networking
to survive a disaster’. 2019 4th International Conference on Information
Systems and Computer Networks, ISCON 2019, Publisher: IEEE, Mathura,
INdia, 10.1109/ISCON47742.2019.9036244; 2019. pp. 250–55.
[48] Kumar A., Krishnamurthi R., Nayyar A., Sharma K., Grover V., Hossain E. ‘A
novel smart healthcare design, simulation, and implementation using health-
care 4.0 processes’. IEEE Access. 2020, vol. 8, pp. 118433–71.
[49] Sundaravadivel P., Kougianos E., Mohanty S.P., Ganapathiraju M.
K. 'Everything you wanted to know about smart health care'. IEEE
Consumer Electronics Magazine, pp. 18-­28, Jan. 2018, Doi: 10.1109/
MCE.2017.2755378. 2018, vol. 7(1).
This page intentionally left blank
 Chapter 15
IoT-­based heart disease prediction system
Rajamohana S P 1, Zbigniew M Wawrzyniak 2, Krishna
Prasath S 3, Shevannth R 3, Raja Kumar I 3,
Mohammed Rafi M 3, and T Hariprasath D 3

In India, almost 80% of patients who die from heart disease do not receive adequate
care. This is a challenging task for doctors because they often seem unable to make
an accurate diagnosis. This condition is extremely expensive to treat. The proposed
solution uses data mining technologies to simplify the decision support system in
order to increase the cost-­effectiveness of therapy. To oversee their patients’ care,
most hospitals use a hospital management system. Unfortunately, many of these
tools do not employ large amounts of clinical data to derive useful information.
Because these systems generate a considerable amount of data in many embodi-
ments, the data is rarely accessed and remains unusable. As a result, making sensible
selections requires a lot of effort during this procedure. The process of diagnosing a
disease currently entails identifying the disease’s numerous symptoms and charac-
teristics. This research employs a number of data mining approaches to assist with
medical diagnostics.

15.1 Introduction
Heart diseases are becoming the leading cause of death [1]. We can use health-­care
data from several hospitals to train a machine learning system to forecast the occur-
rence of health illnesses. Machine learning is highly regarded in health care because
of its ability to process massive datasets faster than humans [2]. Predicting this out-
come would help the doctor plan and deliver better care. It is a structure that uses
a historical cardiac database to predict heart disease. Health-­related signals includ-
ing blood pressure, heart rate, chest discomfort, blood sugar, and cholesterol are
among the input signals. Two additional factors that are also effective in achieving

1
Department of Computer Science & Engineering, Pondicherry University, Karaikal, India
2
Department Electronics and Information Technology, Warsaw University of Technology, Warsaw,
Poland
3
Department of Information Technology, PSG College of Technology, Coimbatore, Tamil Nadu, India
330  Intelligent network design driven by Big Data analytics

Figure 15.1   Types of heart diseases and their symptoms

the desired outcomes are smoking and obesity, both of which are recognized as
significant heart disease symptoms. To obtain the findings, classification techniques
like decision trees and random forests are used along with the data.
The health-­care industry gathers a massive amount of data about health-­care
services. Unfortunately, it often lacks practical information that could aid in making
sound judgments. This work created a revolutionary data mining system capable of
detecting and forecasting cardiac problems.
The outcome of paper demonstrates that each process has its own unique capa-
bility in reaching the stated mining objectives [1]. IHDPS can provide answers to
the complex concerns of what if traditional decision-­making networks could not
use medical files such as age, while blood sugar and blood pressure can predict that
patients will develop heart disease, thereby providing valuable information [3].
As a result, data mining helps to achieve natural evolution in the medical field
[2]. Deep learning is one of the most often utilized methods for the diagnosis of car-
diac problems, along with machine learning and neural networks. The section that
follows provides a brief description of each [3].

15.1.1 Deep learning
The problem in function of heart due to different circumstance is titled as a heart dis-
ease. Some of the familiar heart diseases are congestive heart failure, cardiac arrest,
arrhythmia, stroke, coronary artery disease (CAD), and congenital heart disease as
shown in Figure 15.1. How heart disease varies establish on its type and its symp-
toms are shown in Figure 15.2.
 IoT-­based heart disease prediction system  331

Figure 15.2   Heart disease

The indication to forecast the heart disorder based on the type of heart disorder
symptoms depends on its heart disorder. For example, high blood pressure is one
of the indications for coronary artery for all the people with similar symptom, the
other may differ having the symptom of chest pain. The doctor confirms the heart
disease with the diagnosed report of the patient and several other frameworks. Some
of the most usual heart disorder symptoms and its types are listed in Table 15.1 and
Figure 15.1.

Table 15.1   Heart disease factors with symptoms

Heart disease: risk factors Heart disease: symptoms

Age Discomfort, pressure or heaviness
Sex Pain in the chest or arm, or below the breastbone
Family history Discomfort burning to the back, jaw, throat, or arm
Smoking Fullness, indigestion, or choking feeling
Poor diet Sweating, nausea, vomiting, or dizziness
High blood pressure Extreme weakness, anxiety, or shortness of breath
High blood cholesterol level Rapid or irregular heartbeats
Diabetes
Obesity
Physical inactivity
Stress
Poor hygiene
332  Intelligent network design driven by Big Data analytics

15.2 Related work

Only a few healthcare sectors employ clinical data for predictive purposes, and even
those that do may be limited by a plethora of patient organizational constraints.
Literature surveys are mentioned in Table 15.2. Predictions are based on a doctor’s
intuition rather than extensive research from a scientific database. Incorrect therapy
caused by a faulty diagnosis poses a severe threat to the clinical profession. To solve
these challenges, the following part provided and discussed a data mining strategy
helped by scientific statistics.
Gowrishankar et al. (2017) proposed a framework that estimates and determines
the case’s human body temperature and heartbeat, then redirects the data to the cus-
tomer or server end using a low-­cost microcontroller. It employs three distinct sen-
sors (pulse, heartbeat, and temperature sensors), all of which are heavily controlled
by the microcontroller.
Balamurugan et al. (2017) proposed a framework that uses an Arduino, plus,
and a heartbeat sensor to determine the patient’s pulse and analyze the data to deter-
mine whether the patient is healthy or at risk of having a heart attack [44].
Manisha et al. (2016) proposed a system that assists elderly persons who are
constantly producing heart diseases or heart attacks by tracking their cardiac abnor-
malities and sending an alert alarm to the user’s or the surrounding person’s mobile
phone using a gadget [52]. Additionally, it analyses the heartbeat's data set. It also
makes use of big data analytics, which through utilizing various tools and regula-
tions fosters an environment that is open to customers.
Polu et al. (2019) proposed a system that assists to lower deaths brought by
heart disorder since the main reason of heart disorder deaths is expected due to wait
time and proper treatment. It can be bypassed since the technology will notify the
doctor of his present location and ECG report.
Patel et al. (2014) proposed an article that says angina is very significant for
the sufferer. The sufferer may be saved if additional care and medical assistance are
provided within an hour. To quickly identify heart conditions utilizing the algorithm
and the crucial therapy. A gaming gadget can also serve as lifesaving equipment in
this way. If put into practice, this concept is quite helpful.
Ashrafuzzaman et al. (2013) proposed a method that teaches us how to identify
heart problems by heartbeat and blood data. This technique for identifying heart
issues is highly accurate. With the help of this technique for capturing heart frequen-
cies, we could identify a number of heart-­related conditions. In addition to heart
attacks, it can detect abnormal blood, heart obstruction, and valve circulation. This
is because the heartbeat was captured. It is now capable of identifying heart issues.
Based on heart frequencies, our technology will assist in the early detection of car-
diac issues.
Bo Jin, Chao Che et al. (2018) proposed a neural network-­based model where
the trial and early diagnosis of the heart complaint were performed using electronic
health record (EHR) data from real-­world datasets related to ischemic heart disease
in this research [29]. We used one-­hot encoding and word vectors, which are the
Table 15.2  Related works

Author and Used hardware and Communication

S. No. publication year Proposed framework software components Algorithms used module Limitations
1 Gowrishankar et al. This paper uses sensors LM35, Pulse-­sensor, NA ESP8266 (Wi-­Fi • Data transfer
(2017) [4] to collect the state of Arduino UNO, module) between devices
the heart and body temperature sensor must be fast
temperature with enough with
different test cases minimized delay
and provides an alter response,sensor
message using LCD module should
monitor. extract accurate
readings, two
different sensors
must be used and
should be under
the control of
microcontroller to be
accurate in output.
2 Aboobacker et al. This paper presents a Pulse sensor, Arduino NA Wi-­Fi module • Pulse sensor must
(2017) [5] method to detect UNO detect the pulse
the heart rate of the precisely and
patient, store them, accurately.
and make analysis
based on the sensed
data. It provides
results as if a person
is healthy or not/if
he has a chance of
getting heart attack.
(Continues)
Table 15.2  Continued
Author and Used hardware and Communication
S. No. publication year Proposed framework software components Algorithms used module Limitations
3 Manisha et al. This paper presents a MI Band 2, Android NA Mobile data/Wi-­Fi • Encoder
(2016) [6] method where the phone, big data module must respond
proposed system analytics immediately as the
records heart rate and microcontroller
pulse rate and uses transmits info,
big data analysis to do decoder/decoding
analysis must be accurate
to get an actual
output.
4 Subbalakshmi et al. The paper proposes a ECG monitor, Android NA Mobile data/Wi-­Fi • Wi-­Fi module is
(2020)ok [7] portable ECG monitor phone module used to transfer
where the heart rate and monitor data
will be collected and so connectivity
sent to Android phone. must be constant
In the phone using and fast enough to
detection algorithms, avoid delay. Sensor
abnormality will be module should
recorded and message extract accurate
to ambulance will also readings. Data are
be sent if needed. stored in the server.

(Continues)
Table 15.2  Continued
Author and Used hardware and Communication
S. No. publication year Proposed framework software components Algorithms used module Limitations
5 Ashrafuzzaman et al. This paper proposes Kinect, Xbox one NA Wi-­Fi module • Device detection
(2014) [8] a device “Kinect” of heartbeat must
that monitors heart be precise and
rate and few other accurate otherwise
parameters and detects it will SMS
heart attack. It also triggering alert
sends alert messages about the observer
to emergency contacts will be activated
and makes a Skype and sent to his/her
call to hospital if friends and family.
needed.
6 Ashrafuzzaman et al. This paper says about Smartphone NA Wi-­Fi module • Heart rate
(2013) [9] how to detect blood calculation must be
and heart rate and accurate.
using heart rate • Noise-­free
recoding detect environment is
heart attacks, heart required that is not
blockage, abnormal favourable mostly.
blood, and valve
circulation.
7 Jin et al. (2018) [10] The paper uses sequential Pulse-­sensor, LM35 LSTM ESP8266 Wi-­Fi • Techniques in the
data modeling to temperature sensor, module future to classify
predict heart disorder. Arduino UNO and predict heart
(wearable sensors), diseases.
database • These techniques
maybe less speed
and accuracy.
(Continues)
Table 15.2  Continued
Author and Used hardware and Communication
S. No. publication year Proposed framework software components Algorithms used module Limitations
8 Javeed et al. (2017) The article presents a Heart disease data set Enhanced random NA • Many trees make
[11] solution for by using forest model, the algorithm slow
optimized algorithms. random search and ineffective.
9 Sarangi et al. (2015) The article presents a Heart disease data set Hybrid algorithms NA • The used
[12] solution by using (GA and neural algorithms have too
hybrid techniques. networks) many parameters
for somebody
non expert in data
mining.
10 Mamatha and Shaicy The article presents a Heart disease data set Support vector NA • SVM is not apt for
(2019) [13] solution by using data machine larger data sets.
mining for identify.
11 Bahrami and The article presents a Heart disease data set ICD, KKN NA • Accuracy depends
Shirvani [14] solution by using data on the quality of
mining for identify the data.
using KNN, SVO
classifiers.
12 Bhuvaneswari and The article presents Arduino and heart Naive bayes, NA • Naive Bayes
Kalaiselvi (2012) steps for identify disease data set neural suppose that all
[15] facts based on patient network, features are unique,
reports. decision trees rarely happening in
real time.
13 Subbalakshmi et al. The article presents a Heart disease data set DSHDPS, NA • This algorithm is
(2011) [7] solution for using a Cleveland also notorious as a
Decision Support. Database lousy estimator.
Naive Bayes

(Continues)
Table 15.2  Continued
Author and Used hardware and Communication
S. No. publication year Proposed framework software components Algorithms used module Limitations
14 Jabbar et al. (2011) The article is to learn, Heart disease data set Cluster and NA • They have
[16] develop and identify a association limitation execution
network by implement rule mining time.
associative rules. algorithms
15 Arabasadi et al. The article presents Heart disease data set ANN NA • The greater
(2017) [17] a solution for a computational
proposed hybrid burden, black box
identification model nature, proneness
such as genetic and to overfitting,
ANN algorithm. and the empirical
nature of model
development.
16 Ambekar and The article presents a  Heart disease data set Naïve Bayes, KNN NA • Naive Bayes
Phalnikar (2018) solution for a proposed algorithm assumes that all
[18] disease risk prediction traits are one of
using CNN-­UDRP a kind and that
algorithm. they only occur
infrequently in real
time.

(Continues)
Table 15.2  Continued
Author and Used hardware and Communication
S. No. publication year Proposed framework software components Algorithms used module Limitations
17 Ana and Krishna The article presents a Heart disease data set Naive Bayesian, NA • The multiple
(2017) [19] solution for disease KNN, and Tree algorithms are
prediction for stoke based performed that
patients using increase a rise
wearable sensors in factor of giving
IOT. an improper
information about
the patient.

18 Gupta et al.(2019) This article presents that Heart disease data set KNN, SVM, NA • A real-­time heart
[20] KNN has the best decision tree disease prediction
results comparatively system for
to other algorithms. proactive health
monitoring is not
present.
19 He (2020) [21] Data are collected from Heart disease data set KNN Smart watches • The prediction
wearable IoT devices should be done
for 24 hours and in time series
analysis is made in the manner so the time
cloud. consumption is
relatively high.
20 Shaikh et al. (2015) The article presents a Heart disease data set KNN, Bayesian NA • The system cannot
[22] solution using data and Java Swing handle different
mining technics and is kinds of traits.
implement as a Java
application.

(Continues)
Table 15.2  Continued
Author and Used hardware and Communication
S. No. publication year Proposed framework software components Algorithms used module Limitations
21 Ganesan (2019) [23] In this article, they have Heart disease data set Classification Wi-­Fi Module • Predictions should
presented to diagnose and IoT model be made in a time
heart disease using a series format, since
cloud and IoT-­based this reduces the
model. amount of time
required.
22 Binsalman (2018) This paper proposed a NA Sensors Wi-­Fi module • It is used to monitor
[24] remote monitoring a a pre-­heart disease
heart disease patient. patient only.
23 Bhat (2020) [25] In this paper, they Heart disease data set Convolutional NA • There is a time
have automated the neural network restriction on how
real-­time medical long they can
diagnosis patient. execute.
24 Pooja Arjun (2020) In this article, the author NA Sensors Wi-­Fi module • It is only used to
[26] has proposed an monitor a patient
automated system who is in the early
to monitor the heart stages of cardiac
disease patient. disease.

(Continues)
Table 15.2  Continued
Author and Used hardware and Communication
S. No. publication year Proposed framework software components Algorithms used module Limitations
25 Raju et al. (2022) With the use of Edge-­ Heart disease data set  GSO-­CCNN  NA • The accuracy is
[27] Fog-­Cloud computing, maintained at
the author has the first learning
tried to present a percentages, as
revolutionary smart it is with other
healthcare model in algorithms,
this study. and it requires
greater learning
percentages to
raise it.
 IoT-­based heart disease prediction system  341

core ideas of an extended memory network model, to model the diagnostic events
and read coronary failure events [11]. We usually use the findings to show how cru-
cial it is to value the successional structure of clinical records [7].
Ashir Javeed et al. (2019) proposed an arbitrary timber model and an arbitrary
quest algorithm to identify cardiovascular disease (RSA). This model is intended to
be used in conjunction with a grid quest algorithm.
Srikanta Pattnaik et al. (2015) proposed a cost-­effective model using the genetic
algorithm optimizer technique where the weights were reformed and fed as an input
into the specified network. Ninety percent accuracy was achieved using a hybrid
technique combining GA and neural networks.
Mamatha Alex P et al. (2019) proposed a system that makes use of KNN, ANN,
SVM, and Random Forest techniques. ANN [10] compares all of these data mining
to prognosticate the advanced delicacy of the heart complaint diagnosis.
Boshra Bahrami et al. (2020) proposed a colorful bracket methodology was
used in this study to determine the cause of a cardiovascular problem. Classification
algorithms such as SVO, Decision Tree, and KNN are used to partition the data-
sets [11]. Following the bracket and performance evaluation, the Decision tree is
regarded as the most fashionable option from the dataset for cardiovascular com-
plaint vaccination.
Bhuvaneswari et al. (2018) proposed a system that investigates prior experi-
ence and predicts the level of an object among all objects using the Naive Bayes
classification. The Naive Bayesian and Back Propagation Neural Network catego-
rization algorithms were used in the proposed work [22]. In the supervised learning
environment, the Naive Bayesian classification is utilized to train very effectively,
and the prior backend is formed by Bayesian rules based on the precise structure of
the probability model.
Chinna Rao et al. (2019) proposed a Decision Support in Heart Disorder
Identification Network (DSHDPS) using the nave processing modeling technique
[38]. Heart disease indicators like sex, age, blood pressure, and chest discomfort
can forecast a victim’s likelihood of having a heart issue. Ti functions as an online
application with a questionnaire. Prior to this, the UCI Cleveland database was used
to gather data on heart victims. The nave technique is preferable for identifying
cardiac problems for the reasons listed below: We can obtain a better classification
technique when the amount of data is substantial, the dimensions are distinct from
one another, and the model is compared to other models. Notwithstanding its sim-
plicity, naive algorithms frequently outperform more intricate ones [13].
Jabbar et al. (2019) developed a vaticination system using associative rules and
a novel method that combines the concept of sequence figures and clustering heart
attract vaticination in this study. The first datasheet of heart complaint cases was con-
verted into double format using this method, and the proposed system was applied
to double transitional data. A data set of heart complaint cases with 14 essential
dimensions was extracted from the UCI depository’s Cleveland database. Cluster
Based Association Rule Mining Grounded on Sequence Number (CBARBSN) is
the name of the algorithm. In rule mining, support is a fundamental guideline. To be
included in a commonly used item collection, an item must meet the support criteria.
342  Intelligent network design driven by Big Data analytics

In this investigation, the valid data table is separated based on cluster fractions (dis-
joint subsets of the factual valid table), and each item’s Seq.No and Seq.ID are
predetermined. Frequent item sets have been identified in various clusters based on
Seq.ID, with the most common frequent item set designated as the general item set.
Maximum heart rate > 100, trial blood pressure > 120, old peak > 0, age > 45, and
Thal > 3 indicate a heart attack (frequent item set plant in both clusters in this trial)
[23]. When compared to the preliminary developed system, our proposed algorithm
has a lower prosecution time to mine rules (i.e., 0.879 ms when support is 3), and the
prosecution time changes dramatically as support increases.
Zeinab Arabasadi et al. (2017) proposed a mongrel opinion network for coro-
nary roadway complaints based on the Videl ANN machine literacy algorithm and
inheritable algorithms. The Z-­Alizadeh Sani dataset was used in this study, which
contains 303 case records with 54 features (only 22 were included in the trial),
including 216 cases with coronary artery disease (CAD) [33]. The weights of the
artificial neural network were linked using an inheritable technique first, and then
the ANN model was developed using training data. This sample ANN employs a
feedforward technique with one input and one output subcaste, as well as one hid-
den subcaste with five neurons. The system was evaluated in this trial using a 10-­
fold cross-­confirmation method. We can see from the data that our proposed model
outperformed a simple ANN model in terms of delicacy. We also put our model to
the test in four other well-­known cardiac complaint data sets, with mixed results.
In comparison to an ANN model, our proposed model has a high level of delicacy.
Sayali Ambekar et al. (2018). created a Neural Network technique that predicts
a patient’s disease using a CNN-­UDRP algorithm in structured data. This study
employs a real-­time dataset of cardiac illness. They compared the results of the
KNN and Naive Bayes algorithms and discovered that the accuracy of the NB algo-
rithm is 82%, which is higher than the accuracy of the KNN algorithm. They were
able to forecast disease risk with structured data with a 65 % accuracy [6]. They
were able to achieve accurate disease risk prediction as an output by supplying accu-
rate sickness risk prediction as input, providing us with a greater understanding of
the level of disease risk prediction. The risk of heart disease is classified as low,
high, or medium [12]. Disease risk prediction can now be done in a short amount of
time and at a low cost thanks to this approach [6].
By adjusting the risk factors associated with the disease, ANA et al. (2017)
suggested an ensemble classifier that may be utilized as a general prediction model
for a range of diseases. The system is made up of a microcontroller that is linked to
a number of wearable sensors and the cloud [8]. Sensors record input values, which
are subsequently saved in the cloud and used to generate an alarm message [28].
When a critical level is reached, the system predicts the onset of sickness, allowing
the clinician to take appropriate measures. They investigated the accuracy of various
classification methods as well as ensemble classifiers in this study [8]. The findings
imply that ensemble classifiers are superior to other methods for making predictions.
Anashi Gupta et al. (2019) conducted a result analysis and determined that their
model uses KNN as the training method for classifying individuals with “a chance
of heart disease” and “no chance of heart disease.” In this case, KNN outperforms
 IoT-­based heart disease prediction system  343

well in terms of accuracy, sensitivity, and miss rate [50]. The disadvantage of this
system is that it does not provide a real-­time predictive system that uses sensor data
for proactive health monitoring [7].
He et al. (2020) proposed a framework that transfers client data into client com-
puters and smartphones using Bluetooth, WiFi, and other LAN technologies once
the wearable IoT device has collected client data for 24 hours. The clever will then
upload the data to a distant cloud server, where they will apply a machine learning
algorithm that has already been taught to diagnose the submitted data. They can
attain up to 90% accuracy in this area. However, as the Internet of Things integration
was not tested in this work, speed and power consumption are still unknown.
Shaikh et al. (2015) used Decision Support in a Java Application using Data
Mining Techniques. The system tries to extract hidden knowledge from the data-
base. After determining the probability and in accordance with the probability, pat-
tern matching algorithms will be used to generate the appropriate treatments.
Ganesan et al. (2019) developed an effective cloud and IoT-­based disease diag-
nosis model for monitoring, forecasting, and diagnosing heart disease [24]. The UCI
Repository dataset and medical sensors were utilized to create an effective frame-
work for predicting cardiac disease, which was also investigated [17]. Furthermore,
classification algorithms are utilized to categorize patient data in order to diagnose
cardiovascular disease. The heart disease dataset is used to train the classifier, which
is subsequently employed by the classification method to detect the presence or
absence of heart disease [15]. The trained classifier will evaluate incoming patient
data to determine whether or not the patient has heart disease [27]. Khalid Binsalman
et al. (2018) had used some in healthcare systems, sensor technologies can be ben-
eficial. It is especially beneficial to patients with chronic cardiac disease since it
allows for early intervention, which helps to save lives and cut the risk of death in
half [20]. Especially if these technologies are used in patient-­monitoring systems
through the internet. As a result, using sensor technologies, this research presented a
useful remote monitoring system for heart disease patients [20].
With rising research in the fields of Internet of Things and machine learning,
as well as the growing desire for intelligent and data-­driven healthcare ecosystems,
Bhat et al. (2020) have presented an automatic and real-­time medical diagnosis and
prediction. The suggested strategy in this research seeks to allay these worries. The
interactive, user-­friendly environment that combines real-­time ECG detection and
diagnosis as well as a simple mobile application benefits patients undergoing diag-
nosis. However, accuracy is the key component of a medical application. A system
that is impervious to error overall is produced by deep learning models that can cor-
rectly and automatically learn the required characteristics [5].
V Tamilselvi et al. (2020) proposed a system that may be utilized to constantly,
efficiently, and remotely monitor a patient's temperature and heart rate. The doctor
may remotely monitor the patient's health from anywhere in the globe and provide
consultation based on the results, eliminating the need for the patient to visit the
hospital's O.P.D. (Outpatient Department). Temperature and heart rate thresholds
have been specified. The BLYNK app, which we utilized, sends a notice to the
phone whenever one or both of the readings exceed the threshold. In addition, the
344  Intelligent network design driven by Big Data analytics

suggested system is cost efficient due to the low cost of the components employed.
It's small in size, light in weight, and convenient to transport with the patient wher-
ever they go. The suggested approach is a logical and acceptable method of pro-
viding appropriate help to cardiac patients.[29] developed a system that employs
Edge-­Fog-­Cloud computing to provide a novel smart healthcare paradigm. Data for
this proposed model was collected from a number of hardware sensors. To extract
cardiac properties from signals, peak amplitude, total harmonic distortion, heart
rate, zero crossing rate, entropy, standard deviation, and energy were used. The
characteristics of other attributes were obtained in the same manner, by finding their
"minimum and maximum mean, standard deviation, kurtosis, and skewness." By
improving essential CNN parameters with the CCNN with GSO algorithm [2,] all
of these qualities were introduced to the diagnostic system. By improving essential
CNN parameters with the CCNN with GSO algorithm[2,] all of these qualities were
introduced to the diagnostic system. PSO-­CCNN, GWO-­CCNN, WOA-­CCNN, and
DHOA-­CCNN were 3.7 percent, 3.7 percent, 3.6 percent, 7.6 percent, 67.9 per-
cent, 48.4 percent, 33 percent, 10.9 percent, and 7.6 percent more precise than PSO-­
CCNN, GWO-­CCNN, WOA-­CCNN, and DHOA-­CCNN, respectively. As a result,
the smart healthcare paradigm with IoT-­assisted fog To improve the prediction sys-
tem's accuracy in detecting cardiac illness, the current model could be enhanced in
the future by incorporating more advanced feature selection methods, optimization
approaches, and classification algorithms [2].

15.3 Proposed system

Figure 15.3 represents the block diagram of our proposed system.

Figure 15.3   Block diagram of proposed system

 IoT-­based heart disease prediction system  345

Figure 15.4   Arduino UNO

15.3.1 Arduino UNO

The Arduino UNO [43] uses the ATmega328P microcontroller illustrated in
Figure 15.3. In comparison to other boards, such as the Arduino Mega, it is simple
to use. The board consists of digital and analog input/output (I/O) pins, shields, and
other circuitry [4]. The Arduino UNO contains six analog input pins, fourteen digital
pins, a USB connection, a power jack, and an ICSP (In-­Circuit Serial Programming)
header. It’s written in the IDE programming language (Integrated Development
Environment). It is applicable both online and offline [30]. The block diagram of
our suggested system is shown in Figure 15.4. The hardware requirements are stated
below.
The major components of Arduino UNO board are as follows:

• USB connector
• power port
• microcontroller
• analog input pins
• digital pins
• reset switch
• crystal oscillator
• USB interface chip
• TX RX LEDs
346  Intelligent network design driven by Big Data analytics

Figure 15.5   Heartbeat sensor

15.3.2 Heartbeat sensor
The sound of a person’s heartbeat is formed by the contraction or expansion of the
valves in his or her heart as they force blood from one area to another. The rate of
the heartbeat is measured in beats per minute (BPM), and the pulse is the heartbeat
sensed in any artery adjacent to the skin [1].
The heartbeat sensor shown in Figure 15.5 is powered by the photoplethysmog-
raphy principle. It detects changes in blood volume traveling through any organ of
the body, resulting in a change in the intensity of light passing through that organ
(vascular region) [1]. In applications that measure heart pulse rate, pulse timing is
very crucial. The amount of blood that flows is determined by the rate of heartbeats
and blood absorption.
A basic heartbeat sensor consists of a light-­emitting diode and a detector, such
as a light-­detecting resistor or a photodiode. As a result of the heartbeat pulses, the
flow of blood to various parts of the body varies [10]. When a light source, such as a
led, illuminates tissue, the light either reflects (as in finger tissue) or transmits (as in
eye tissue) (earlobe). Some of the light is absorbed by the blood, and the light detec-
tor detects the transmitted or reflected light [16]. The blood volume of the tissue
determines the amount of light absorbed [16]. The detector produces an electrical
signal proportional to the heartbeat rate [34].
 IoT-­based heart disease prediction system  347

Figure 15.6   Temperature sensor

15.3.3 Temperature sensor
The LM35 series, as illustrated in (Figure 15.6), are precision integrated-­circuit
temperature devices with a linearly proportional output voltage to temperature in
degrees Celsius [10]. Unlike linear temperature sensors calibrated in Kelvin, the
LM35 temperature sensor has the advantage of not requiring a sizable constant volt-
age to be subtracted from the output in order to provide appropriate Centigrade
scaling [46].

15.3.4 Pressure sensor
Figure 15.7 shows how a constant-­area sensing device in pressure transducers
responds to the force generated by fluid pressure. The applied force will cause the
diaphragm of the pressure transducer to deflect. The internal diaphragm deflection is
detected and transformed into an electrical output [18]. Microprocessors, program-
mable controllers, computers, and other electronic equipment can therefore monitor
pressure [14].

15.3.5 Liquid crystal display (LCD) display (16 × 2)

As shown in Figure 15.8 an LCD 16*2 is a sort of electronic display device that
shows data and messages. It shows 32 characters (16*2=32), each made up of 588
(40) Pixel Dots. It has 16 columns and 2 rows, as the name says. As a result, rather
than 1,280 [3] the total number of pixels in this LCD is 32 ×40.
The proposed method detects a heart attack by monitoring the heartbeat through
the object’s internet connection. A heartbeat sensor [30], a Wi-­Fi module, and
an Arduino board are used in our solution [51]. The heartbeat sensor will begin
to detect heartbeat readings after the programme is set, and the human heartbeat
will be displayed on the LCD screen [49]. A Wi-­Fi module can also be used to
348  Intelligent network design driven by Big Data analytics

Figure 15.7   Pressure sensor

Figure 15.8   LCD display

 IoT-­based heart disease prediction system  349

Figure 15.9   Flow diagram

communicate data over the Internet [31]. The device can help determine a person’s
health by measuring their heart rate and comparing it to a set point [32]. After these
constraints are set, the system will begin monitoring the patient’s heart rate, and if
the heart rate goes over or below the stated limit, the system will alert the user. For
this work, we’re using an Android app model that will measure a patient’s heart rate,
monitor it, and deliver an urgent message about the risk of a heart attack [33]. We
utilize Deep Learning architecture since it is an artificial neural network (ANN) that
allows us to manage the stream by combining Inputs according to trained Weights.
As a result, I’m calling for more adaptability in controlling outputs [34]. Using Pre-­
historic data from the heart disease prediction database and simple ANN, we would
create a Deep Learning model. When new data is added to the database, the model
should try and predict cardiac disease, update the database, and notify the physician
[35].
Figure 15.9 presents the circuit diagram of the proposed system [36].
As shown in (Figure 15.10), the proposed network uses an arduino and the sen-
sors which we use are temperature (LM05), heartbeat, and pressure (MPX4115)
sensors [37]. It also has a com port and a terminal connected so by which help us to
detect and use the output from the sensors, The Deep learning model is trained with
an accuracy of 86.18%, the data which we acquire from the sensors and fed into the
Deep learning model and results in an acquired respectively. They also tried training
with some machine learning algorithms which we witness an accuracy of 74.3% for
both naïve Bayes and logistic regression [38]. Our system works in real-­time and the
number of sensors deployed is comparatively less, so the accuracy can be increased
by adding more sensors and increasing the volume of trained data used to train our
model [39].
350  Intelligent network design driven by Big Data analytics

Figure 15.10   Circuit diagram

15.4 Advantages of proposed system

• The accuracy of the system is high.

• The system also tries to detect all the health anomalies of the user.
• The system does not detect heart rates alone but also BP of the victim using
pulse rate and body.

15.5 Limitations of proposed system

• Network failure may lead to failure of alerting systems.

• The system uses Bluetooth and the power consumption will be a bit high.
• Power consumption and performance of the system is still unknown.
• The volume of data taken to train is relatively less.

15.6 Results and discussion

The final output of the model is displayed under two separate sections as the sens-
ing section and the predicting section. The results of heart rate, and temperatureare
displayed in LCD 16*2 displays as shown Figure 15.11.
The results of predicting section are shown in Figure 15.12. If the model detects
chance of heart disease based on the sensing results, it prints “chance of Heart
Disease,” else it prints “Normally Functioning Chance of Heart Disease if Less.”
 IoT-­based heart disease prediction system  351

Figure 15.11   Sensing results

15.7 Conclusion

This paper provides a comprehensive study on heart disease diagnosis using a per-
son’s heartbeat in the proposed system, the sensor [40] detects heartbeat signals and
sends them to the Arduino UNO. The Arduino [41], in conjunction with the WiFi
module, sends those signals to the cloud, where they are analyzed and predicted
using a deep learning model. The Android application [42] receives an appropriate
notification. Because we use artificial neural networks to predict the results, the
accuracy is higher. Even though we have improved the accuracy of the predicted
results, the network has some flaws. The system functions by sending a heartbeat
to a Wi-­Fi module linked to a microcontroller [43]. We limited the system’s ability
to measure heartbeats. In the future we can store all the data in Edge devices and
perform further analytics and add the SMS and e-­mail module to notify the patients
and doctors through our system [44].

Figure 15.12   Predicting results

352  Intelligent network design driven by Big Data analytics

References

[1] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[2] Sudhakaran S., Kumar S., Ranjan P., et al. ‘Blockchain-­based transparent
and secure decentralized algorithm’. International Conference on Intelligent
Computing and Smart Communication 2019. Algorithms for Intelligent
Systems. Springer; 2020.
[3] Sarangi L., Mohanty M.N., Pattnaik S., Shubham M. ‘An intelligent deci-
sion support system for cardiac disease detection’. International Journal Of
Control Theory And Applications, International Science Press. 2015, vol.
8(5), pp. 2137–43.
[4] Gowrishankar S., Prachita M.Y., Prakash A. ‘‘IoT based heart attack detec-
tion, heart rate and temperature monitor, international journal of computer
applications, (0975 – 8887), volume 170(5)’.2017.
[5] Aboobacker A., Balamurugan D.S. ‘Heartbeat sensing and heart attack de-
tection using internet of things: iot’. International Journal of Engineering
Science and Computing, 6662-­6666, Volume 7 Issue No.4, April 2017.
2017.
[6] Manisha M., Sindhura V., Ramaya P., Neeraja K. ‘IoT on heart attack de-
tection and heart rate monitoring’. International Journal of Innovation in
Engineering and Technology. 2016, vol. 196.
[7] Ramesh K., Rao M.C., Subbalakshmi G ‘Decision support in heart disease
prediction system using naive bayes’. Indian Journal of Computer Science
and Engineering (0976-­5166). 2020, vol. 2(2), pp. 170–76.
[8] Patel S Shivam, Chauhan Y ‘Heart attack detection and medical attention us-
ing motion sensing device- kinect’. International Journal of Scientific and
Research Publication (2250-­3153). 2019, vol. 4.
[9] Ashrafuzzaman M.: ‘Heart attack detection using smart phone’. International
Journal of Technology Enhancements and Emerging Engineering Research 1.
2020, pp. 23–27.
[10] Jin B., Che C., Liu Z., Zhang S., Yin X., Wei A.X. ‘Predicting the risk of
heart failure with EHR sequential data modeling’. IEEE ACCESS: Practical
Innovations, Open Solution, PP. 1-­1, 0.1109/ACCESS.2017.2789324.. 2018.
[11] Javeed A., Zhou S., Yongjian L, et al. ‘An intelligent learning system based
on random search algorithm and optimized random forest model for im-
proved heart disease detection’. IEEE ACCESS: Practical Innovations,
Open Solutions, Vol. 7, Pp. 180235-­180243, 2019, Doi: 10.1109/
ACCESS.2019.2952107. 2019.
[12] Sarangi L., Mohanty M.N., Pattnaik S.: ‘An intelligent decision support sys-
tem for cardiac disease detection’. International Journal of Control Theory
and Applications. 2019, vol. 8(5), pp. 2137–43.
 IoT-­based heart disease prediction system  353

[13] Mamatha Alex P., Shaji S.P. ‘Prediction and diagnosis of heart disease pa-
tients using data mining technique’. 2019 International Conference on
Communication and Signal Processing (ICCSP), Publisher: IEEE, 2019, pp.
0848-­0852, doi: 10.1109/ICCSP.2019.8697977; 2019.
[14] Bahrami B., Shirvani M.H. ‘Prediction and diagnosis of heart disease by data
mining techniques’. Journal of Multidisciplinary Engineering Science and
Technology (JMEST). 2020, vol. 2(2),(3159-­0040).
[15] Bhuvaneswari R., Kalaiselvi K. ‘Naïve Bayesian classification approach in
healthcare applications’. International Journal of Computer Science and
Telecommunication. 2018, vol. 3(1), pp. 106–12.
[16] Chandra D.P., Deekshatulu B.L., Jabbar M.A ‘Cluster based association
rule mining for heart attack prediction’. Journal of Theoretical and Applied
Information Technology, 196-­201, (1992-­8645). 2019, vol. 32(2).
[17] Arabasadi Z., Alizadehsani R., Roshanzamir M. ‘Computer aided decision
making for heart disease detection using hybrid neural network-­genetic algo-
rithm’. Computer methods and programs in biomedicine. 2017, vol. 141, pp.
19–26.
[18] Amebekar S., Plalnikar R. ‘Disease risk prediction by using convolution neu-
ral networks’. IEEE, 2018 Fourth International Conference on Computing
Communication Control and Automation (ICCUBEA), 2018, Pp. 1-­5, Doi:
10.1109/ICCUBEA.2018.8697423. 2018.
[19] Ana R., Krishna S. ‘IoT based patient monitoring and diagnostic prediction
tool using ensemble classifier’. IEEE, 2017 International Conference on
Advances in Computing, Communications and Informatics (ICACCI), 2017,
Pp. 1588-­1593, Doi: 10.1109/ICACCI.2017.8126068. 2017.
[20] Gupta A., Yadav S., Shahid S. ‘HeartCare: iot based heart disease predic-
tion system’. Presented at 2019 International Conference on Information
Technology (ICIT), 88-­93, 2019;
[21] Qhe, Maag: ‘Heart disease monitoring and predicting by using machine learn-
ing based on iot technology’. UTC from IEEE Xplore. May 28, 2021.
[22] Shaikh S., Sawant A., Paradkar S., Patil K Electronic recording system-­heart
disease prediction system 2015 International Conference on Technologies for
Sustainable Development (ICTSD; Mumbai, India, 2015.
[23] M.G ‘IoT based heart disease prediction and diagnosis model for healthcare
using machine learning models’. IEEE, 2019 IEEE International Conference
on System, Computation, Automation and Networking (ICSCAN), 2019, Pp.
1-­5, Doi: 10.1109/ICSCAN.2019.8878850. 2019.
[24] BinSalman K., Fayoumi A. ‘Effective remote monitoring system for heart dis-
ease patients’. 2018 IEEE 20th Conference on Business Informatics; Vienna,
Austria; 2018.
[25] Bhat T., Akanksha, Shrikara, Bhat S., T M A real-­time iot based arrhyth-
mia classifier using convolutional neural networks 2020 IEEE International
Conference on Distributed Computing, VLSI, Electrical Circuits and Robotics
(DISCOVER; Udupi, India, 2020. https://ieeexplore.ieee.org/xpl/mostRecen-
tIssue.jsp?punumber=9278009
354  Intelligent network design driven by Big Data analytics

[26] Baad P.A. ‘IOT based health monitoring system using Arduino’. Journal of
Engineering, Computing and Architecture. 2020, vol. 10(5).
[27] Raju K.B., Dara S., Vidyarthi A., Gupta V.M., Khan B Smart heart disease
prediction system with iot and FOG computing sectors enabled by cascaded
deep learning model’.Computational Intelligence and Neuroscience. 2022,
vol. 2022(5), 1070697.
[28] Kumar S. Evolution of Software-­Defined Networking Foundations for IoT
and 5G Mobile Networks. IGI Publisher; 2020. p. 350.
[29] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[30] Reghu S., Kumar S. ‘Development of robust infrastructure in networking
to survive a disaster’. 2019 4th International Conference on Information
Systems and Computer Networks, ISCON 2019; Mathura, India; 2019. pp.
250–5.
[31] Kumar S., Cengiz K., Vimal S., Suresh A. ‘Energy efficient resource migra-
tion based load balance mechanism for high traffic applications iot’. Wireless
Personal Communications. 2021, vol. 10(3), pp. 1–19.
[32] Kumar S., Cengiz K., Trivedi C.M., et al. ‘DEMO enterprise ontology with
a stochastic approach based on partially observable Markov model for data
aggregation and communication in intelligent sensor networks’. Wireless
Personal Communication. 2022.
[33] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic Torus
Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering (IJITEE). 2019, vol. 8(6), pp.
2278–3075.
[34] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[35] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring
using machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya
M.N., Satapathy S.C. (eds.). Intelligent Manufacturing and Energy
Sustainability. Smart Innovation, Systems and Technologies. 265. Singapore:
Springer; 2022.
[36] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[37] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wire-
less sensor network’. Proceedings - 2015 5th International Conference on
Communication Systems and Network Technologies CSNT; Gwalior, India;
2015. pp. 194–200.
 IoT-­based heart disease prediction system  355

[38] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distrib-

uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India; 2014.
[39] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, Internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India; 2021. pp. 01–6.
[40] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium. 2015, vol. 2015, pp. 2363–7.
[41] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’.
Proceedings - 2015 International Conference on Computational Intelligence
and Communication Networks, CICN 2015; Jabalpur, India; 2016. pp. 79–84.
[42] Sharma A., Awasthi Y., Kumar S. ‘The role of Blockchain, AI and IoT for smart
road traffic management system’. Proceedings - 2020 IEEE India Council
International Subsections Conference, INDISCON 2020; Visakhapatnam,
India; 2020. pp. 289–96.
[43] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of Indian roads’. Proceedings of the
Confluence 2020 - 10th International Conference on Cloud Computing, Data
Science and Engineering; Noida, India; 2020. pp. 63–76.
[44] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation of
fault tolerance technique for Internet of things (IoT)’. Proceedings - 2020 12th
International Conference on Computational Intelligence and Communication
Networks, CICN 2020; Bhimtal, India; 2020. pp. 154–9.
This page intentionally left blank
 Chapter 16
DIAIF: Detection of Interest Flooding using
Artificial Intelligence-­based Framework in
NDN android
P Vimala Rani, Rajkumar 1, Narshimha Malika Arjunan 1,
and S Mercy Shalinie 1

In today’s world, information-­centric networking (ICN) is a brand-­new next-­generation

network for distributing multimedia content. ICN focuses on sharing content across the
network rather than obtaining content from a single fixed server [1]. In-­network cach-
ing aids in the dissemination of content from the network, and the ICN also includes
a number of intrusive security mechanisms. Despite the ICN network’s many security
measures, several attacks, especially interest flooding attacks (IFA), continue to wreak
havoc on the network’s distribution capability. In order to address security threats,
the literature includes a number of mitigating procedures. However, legitimate users’
requests are misclassified as an attack in an emergency circumstance, affecting the net-
work’s QoS [2]. In this chapter, Detection of Interest Flooding Attack using Artificial
Intelligence Framework (DIAIF) is proposed in ICN. DIAIF seeks to lighten the load
on ICN routers by removing the source of the attack without interfering with legitimate
user requests. DIAIF depends on router feedback to assign a beneficial value (BV) to
each piece of content and to block dangerous users based on the BV. The ICN test-
bed was designed to assess the proposed DIAIF’s performance in terms of QoS during
severe flood scenarios, responding with malicious content without interfering with gen-
uine user requests, and identifying the source of attack in a communication scenario.

16.1   Introduction

The distributed denial-­of-­service (DDoS) is a risky attack that can severely decrease
network performance by up to 99 per cent. This attack infected the network with a
script named Trin00, which caused the machines to crash severely, and it began on
July 22, 1999 [3]. Despite the fact that two decades have passed, Internet security
specialists continue to face significant challenges. According to a Kaspersky survey,

1
Department of Computer Science and Engineering, Thiagarajar College of Engineering, Madurai, India
358  Intelligent network design driven by Big Data analytics

Figure 16.1   Benefits of ICN on application

DDoS attacks have climbed by up to 18 per cent from the year 2021 to today [4].
Similarly, in 2016, Mirai, a botnet that targets 200–300 Internet of Things (IoT)
devices and also diverts Domain name system (DNS), creating massive distribution in
content distribution, was discovered. GitHub was hit by the greatest enormous attack
ever, with 1.3 terabits of bandwidth exploding and bringing the network to its knees.
This is due to transmission control protocol (TCP)/ Internet protocols (IP’s) lack of
built-­in security, which causes more collateral harm to the Internet architecture. Many
clean slate architectures have been proposed to defend the network against attackers
and inefficiency in providing security of today’s Internet architecture of TCP/IP, with
the information-­centric network being the most nominated by the Internet security
community in recent years. ICN has built-­in features that help you communicate with
multimedia content flawlessly. Content store (CS), pending interest table (PIT), and
forwarding information base (FIB) are some of the functionalities available. ICN is
concentrating on content security rather than end-­host security. This built-­in secu-
rity enhancement reduces the danger of security flaws that come with the traditional
TCP/IP approach. The in-­built components of ICN that make the content distribution
process easier are shown in Figure 16.1. Several security solutions for IFA have been
identified in the literature [5, 6]. The named data networking (NDN) simulations were
used to implement the majority of the security identification and solutions. When
attacks are developed in real-­time scenarios, however, the simulation results will not
adapt. As a result, we propose the ICN testbed in conjunction with NDN functions in
order to detect real-­time threats using artificial intelligence techniques.

16.2   Background

In this section, the ICN communication model and its components are described in
Section 16.2.1, and attack scenario is described in Section 16.2.2.
 DIAIF  359

Figure 16.2   ICN interest forwarding process

16.2.1   ICN communication model

ICN’s material is distributed efficiently using two types of packets: interest and data
packets. Each router is made up of three parts: CS, PIT, and FIB. When the router
receives the interest packet from the consumers, it first checks the content from CS.
If not, it looks up the entry in PIT. If an entry is accessible, it is forwarded to FIB;
otherwise, it is added to PIT as a new entry. FIB is in-­charge of finding content from
the nearest router with choosing multiple next-­hop entries. If content is detected on
the downstream node, the packet is sent to the user using the PIT entry. Because PIT
contains a name for the interface and a user interest packet. Meanwhile, the data
packet is saved in CS only if the PIT entry for the same interest packet has more
entries. If matching or next-­hop address is not found for the interest packet, the
router returns the negative acknowledgement (NACK) to consumers. Figure  16.2
shows the ICN interest forwarding model.

16.2.2   Generating IFA in NDN android

The IFA is caused by malevolent users who deliberately overload the network with
content. The example of IFA in NDN android is shown in Figure 16.3. When a router
is overburdened, genuine users’ packets are dropped. When the router’s PIT is occu-
pied by IFA, the PIT’s storage capacity is depleted, and legitimate users’ packets are
not stored in the PI [7]. Malicious users, on the other side, employ ICN routers to
damage the performance of data producers. ICN components are the umbrella under
which NDN android operates. The emergency user requests are dropped when IFA is
generated in the NDN android service, resulting in a No Route or NACK acknowl-
edgment. With false interest packets, a malicious user sends interest with nonexist-
ent content (now unavailable in the network). PIT will be filled as a result of this
fake interest packet, and this packet will remain for a longer time until PIT identifies
the relevant data packet [8].
360  Intelligent network design driven by Big Data analytics

Figure 16.3   IFA on NDN android

16.3   Proposed methodology: RD Iterative Adaptive Inverse

Filtering (RD-IAIF)

In this section, Real-­time Detection of Interest Flooding Attack using Artificial

Intelligence Framework for NDN multimedia application is presented. RD-­IAIF is
capable of identifying a number of attack possibilities in NDN, despite the fact that
IFA is managing the network by sending significant amounts of interest, which led
to the inclusion of the IFA in this study. The IFA generation is mentioned in Section
16.2.2. Before detecting the IFA, the data are collected from the NDN android dur-
ing the time of attack. The parameters are helpful in the identification of attack that
is described as follows:
Packet count: It refers that the count of the same interest packet is received
from the total number of content that is received during the time t.
j
†k=1 n(Ik )
(I) = j (16.1)
‍ †r=1 n(Ir ) ‍

where ‍n(Ik )‍is defined as the sum of the number of particular requests for I. ‍n(Ir )‍is
defined as a count of requests that is received by the router.
Variance of packet count: It is the measurement of comparing the relative
packet count influence the content distribution of ICN.
PN
(Ik  NI)
S2 (I) = k=1 (16.2)
‍ N1 ‍
P
where ‍ NI ‍defines the mean of same interest packet ‍NI = 1n Nk=1 Ik .‍
 DIAIF  361

Standard deviation of inter-­arrival interest packet: It is the measurement of

how inter-­arrival of the same interest packet is deviated from the mean.
s
P N
 (Ik ) = N1 (Ik  NI) (16.3)
‍ k=1
‍
EWMA: EWMA stands for exponential weighted moving average. It finds the like-
lihood ratio of same content request in ICN. It is useful to find how the particular
content being overruled in time window t. It adds the weight to a content and weight
on the past content objects is decreased exponentially. It increases the ratio of a
recent data by rejecting the older one. The weigh factor is assigned between 0 and 1.
‍ C(t) =˛.y(t) + (1  ˛)  C(t  1)‍ (16.4)
Satisfaction ratio (‍ ‍): It is the ratio of satisfied interest packet from the total packet
that is received by the router. During the time of IFA, the satisfaction ratio will be
reduced.
Xn
NDk (t)
‡ (rjl , ti ) = (16.5)
‍ k=0
NIk (t) ‍

where ‍NDk ‍and ‍NIk ‍are number of data and interest packets belong to ‍jth‍interface of
j
router ‍(rl )‍with respect to time ‍(t)‍.
PIT utilization (‍ ‍): PIT entries are calculated at each time cycle (t).
Pn
k=0 Nentriesk (t)
(rl , ti ) = (16.6)
‍ Nsize ‍

16.3.1 Attack detection
The Multi Feature(MF)-­based Adaptive Neuro-­Fuzzy Inference System (ANFIS)-­is
a attack detection technique on NDN android. The Takagi–Sugeno system is used
to model the proposed multi function adaptive neuro-­fuzzy inference system (MF-­
ANFIS), where input is traveled through the multiple layers and output is linear
combination weighted average. Figure 16.4 shows the proposed MF-­ANFIS model.
The sample of Takagi–Sugeno system with two rule set is as follows:
‍ If x is A1 and y is B1 then f1 = p1 x + q1 y + r1‍ (16.7)
‍ If x is A2 and y is B2 then f2 = p2 x + q2 y + r2‍ (16.8)
where ‍Ai , Bi ‍are the fuzzy set and ‍pi , qi , ri ‍are the linearly independent output vari-
ables. ANFIS contains five layers to process the input to output and nine if-­then
rules as follows:
Layer 1: All the square input nodes are obtained by the parameters from the
whole data stream. The membership function to input node is given in the following
equation:
‍ O1,i = Ai (x), i 2 1, 2, 3; O1,i = Bi3 (y), i 2 1, 2, 3‍ (16.9)
362  Intelligent network design driven by Big Data analytics

Figure 16.4   List nearest FaceID

Layer 2: In this layer, every node is fixed and it multiplied the product of the input
to obtain the output values.
‍ O2,i = wi = Ai (x)  Bi3 (y), i 2 1, 2, 3...9‍ (16.10)
Layer 3: Every node in this layer is considered to be fixed and it does the normaliza-
tion function.
wi
O3,i = wi = w +w +w , i 2 1, 2, 3...9 (16.11)
‍ 1 2 3 +...w9 ‍
Layer 4: All square nodes are adaptive. The output of this node is considered to be
normalized values.
‍ O4,i = wi .fi = wi .(pi x + qi y + ri ), i 2 1, 2, 3...9‍ (16.12)
Layer 5: It sums all incoming single and produces the proper output. All square
nodes are adaptive. The output of this node is considered to be normalized values.
P
X w i fi
wi fi = P , i 2 1, 2, 3...9. (16.13)
i
O5,i = f =
‍ i i ‍

16.4   Real-time deployment on emergency applications

The NDN Forwarding Daemon (NFD) is a key component of ICN’s content dis-
tribution and update provisioning. NFD supports a variety of platforms, including
Linux, FreeBSD, Mac OSX, Android, and Raspberry Pi and allows named data
 DIAIF  363

communication instead of IP transmission [9]. To facilitate interest/data communi-

cation, NFD offers a variety of management protocols. PIT and FIB face lists, as well
as removal entries from PIT, CS, and FIB, are all included in the NFD structures.
When the router PIT is full, it allows each router to share the notice as “NACK.”
NFD supports a number of routing protocols that should help with IP over NDN
migration.

16.4.1   Communication establishment through NFD

Figure  16.5 shows the usage of NDN android on various enterprises in real-­
time. In NDN applications, each producer delivers a data packet to the routers
with a name prefix. As a result, if the data packet’s name prefix is frequently
requested, each router keeps it in CS. On the other hand, if customers request
the same name prefix, it is immediately shared with them. The NDN forwarders
can provide you with the data packet. The following methods can be used to cre-
ate NDN data packets: Consumer router configuration, prefix announcements,

Figure 16.5   IFA generation and detection in ICN testbed

364  Intelligent network design driven by Big Data analytics

Figure 16.6   IFA generation and detection in ICN testbed

and data discovery tactics are all done manually. The NFD approach with cache
for the NDN application is shown in Figure 16.6. The router caches the content
when two users request the same prefix name, as shown in step 9 of Figure
16.6. The user of the NDN app can request any type of data to be forwarded to
the nearest network. If the content is not available on the local router, it is auto-
matically forwarded to the producer via the gateway router. The FIB is updated
for each and every transaction. The local face and name prefix of user app1 are
first registered with the local NFD router. Users’ names and interfaces are then
saved in the local NFD. As a result, the content itself, rather than the location
information, is routed to the gateway router. In this method, NDN achieves
name-­based routing rather than location-­based routing.

16.4.2   NDN application on android

The NDN application on android is feasible solution for emergency situations. NFD
source code is archived from the GitHub to enable all packages and faces. The NFD
allows user to create a face, modify, and delete whenever they want. The NFD syn-
tax for creating face are listed below:
Face creation
 DIAIF  365

Figure 16.7   List nearest FaceID

• nfdc face create[remote] <  FACEURI > [[persistency] < PERSISTENCY > ]
[local < FACEURI >]
• nfdc face create remote udp://​education.​example.​com

Face creation with remote URI

• nfdc face create[remote] <  FACEURI > [[persistency] < PERSISTENCY > ]
[local < FACEURI >]
• nfdc face create remote ether://[07:00:13:02:02:02] local dev://eth4 persistency
permanent

Figure 16.8   User interface on NFD app

366  Intelligent network design driven by Big Data analytics

Figure 16.9   User chat on NFD app

Congestion-­threshold: It creates a face with a congestion marker. [congestion –

]marking – interval – INTERVAL >]
nfdc face create remote udp://​education.​example.​com congestion-­marking-­
interval 100 default-­congestion-­threshold 65,536
Remove face
It allows the user to remove the face.
nfdc face destroy [face]< FACID|FACEURI >
nfdc face destroy 300
NFD route list: The route list contains nexthop address and prefix name in FIB.
The commands are as follows:
Route list

• nfdc route[list[[nexthop]< FACID|FACEURI >][origin < ORIGIN >]]

• nfdc route list nexthop 421

Add route list

• nfdc route add [prefix] < PREFIX > [nexthop] <  FACID|FACEURI> [origin <
ORIGIN >]
• nfdc route show prefix /localhost/nfd

Add new route list

 DIAIF  367

Figure 16.10   IFA on NFD app

• nfdc route add [prefix] < PREFIX > [nexthop]< FACEID|FACEURI >  [origin
< ORIGIN >]
• nfdc route add prefix / nexthop udp://​router.​example.​net

Remove route list

• nfdc route remove [prefix]< PREFIX> [nexthop] < FACID|FACEURI > [origin
< ORIGIN>]
• nfdc route remove prefix /ndn nexthop 421 origin static

NFD face is created with nexthop and remote URI for request and response of
packets. Figure 16.7 shows the face ID of NDN remote URI. Figure 16.8 shows the
updated NDN components after creating the face ID and route list.
We put NFD to the test on two android devices with different compatible mod-
els and configurations. NFD tools are also used to configure the producer device.
When the gateway router does not have the packet in cache for each test, it sends the
request to the producer. When name-­based routing is used in the network, the CPU
368  Intelligent network design driven by Big Data analytics

Figure 16.11   User chat on NFD app

use is lowered. Each customer request originates from a 100 MB android handset.
The test is run on two android phones: a Samsung phone with 4.0 GB RAM and a
Hisilicon Kirin 710 CPU, and a Motorola phone with a 2.1 GHz quad-­core CPU.

16.4.3 Attack detection on NFD android using AI

The NDN users can share their information with each other on android NDN.
Figure 16.9 shows the sample chat model in NDN.
The transparent content dissemination strategy leads to DDoS attack on NDN
application. This DDoS attack on NDN platform is considered to be IFA. . It is
partially different from the TCP/IP DDoS attack. Once the IFA is generated from
the third party, the content dissemination of NDN platform is disturbed and it leads
to “No Route” and NACK notification issues [10]. Figure 16.10 shows the IFA is
generated on android NDN. The attacker sends the 1,000/s to ruin the NDN applica-
tion running process in the communication. MF-­ANFIS detects the attack in timely
manner that helps to design the NDN communication process more secure. It mini-
mizes the training error between the desired output and actual input. The input of
MF-­ANFIS is derived from Section 16.3. MF-­ANFIS is run on the MATLAB® ver-
sion 2015a.

Figure 16.12   User chat on NFD app

 DIAIF  369

Figure 16.13   CPU consumption on NFD app

The six parameters are given as input to the MF-­ANFIS model and it produces
one output. Before being fed into the AI model, it must be normalized to [0..1].
In MF-­ANFIS gaussmf membership, functions are used to obtain the high accu-
racy results. The center of gravity functions is used as a defuzzification model. MF-­
ANFIS uses if-­then rules for decision making on android NFD. Figure 16.11 shows
the proposed detection model with multiple features. Figure 16.12 shows the MF-­
ANFIS decision making on NDN android.
During this heavy flooding period, the CPU consumption is considerably high.
When a user wants to request or obtain the data from the NDN android, the pack-
ets are lost due to the heavy load in the CPU process. With the help of MF-­FCM,
the CPU load is captured. Figure 16.13 shows the CPU load during the IFA. The
“green” line indicates that the traffic has arrived to the network. The “blue” line
represents the outbound traffic that is nothing but the router service the packet to the
destination.

16.5 Conclusion

In the event of an emergency, the NDN on android provides a far faster response
time. The NDN application is vulnerable to a severe attack due to transparent for-
warding table updates and name-­based routing. During a major flood, the NDN
app fails to consider the incoming packet, resulting in significant packet loss in an
emergency circumstance. The artificial intelligence-­based MF-­ANFIS is presented
to detect the attack pattern and frequency of the request pattern. MF-­ANFIS detects
attacks and assists in the discovery of malicious content patterns. As a result, IFA
detection is required for real-­time applications, and additional research into NDN
security challenges pave the path for a healthy and intelligent future.

References

[1] Antonakakis M., April T., Bailey M., et al. ‘Understanding the mirai botnet’.
26th USENIX Security Symposium(USENIX Security 17); Vancouver, BC;
2017. pp. 1093–110.
[2] Jose S. ‘Global mobile data traffic forecast update, 2016–2021 white paper,
index’. Cisco Visual Networking. 2017, p. 180.
370  Intelligent network design driven by Big Data analytics

[3] Dutta N. ‘Introduction to Information-­Centric Networks. Information Centric

Networks (ICN Springer, Cham)’. Osterweil 2019, 20 years of ddos: a call to
action, 2019, arXiv preprint arXiv:1904.02739. 2021, pp. 1–25.
[4] Report finds 18% rise in ddos attacks in Q2 2019
. 2019. Available from https://usa.kaspersky.com/about/press-releases/2019_​
report-finds-18-percent-rise-in-ddos-attacks-in-q2-2019
[5] Khelifi H., Luo S., Nour B. ‘Named data networking in vehicular AD hoc
networks: state-­of-­the-­art and challenges’. IEEE Commun Surv. Tutor. 2019,
pp. 320–51.
[6] Nour B., Sharif K., Li F., Wang Y. ‘Security and privacy challenges in
information-­centric wireless Internet of things networks’. IEEE Secur.
Privacy. 2019, pp. 5–45.
[7] Fang C., Yao H., Wang Z., Wu W., Jin X., Yu F.R. ‘A survey of mobile
Information-­Centric networking: research issues and challenges’. IEEE
Communications Surveys & Tutorials. 2018, vol. 20(3), pp. 2353–71.
[8] Bouk S.H., Ahmed S.H., Kim D., Park K.-J., Eun Y., Lloret J. ‘LAPEL: hop
limit based adaptive pit entry lifetime for vehicular named data networks’.
IEEE Transactions on Vehicular Technology. 2018, vol. 67(7), pp. 5546–57.
[9] Zhang H., Li Y., Zhang Z., Afanasyev A., Zhang L. ‘Ndn host model’. ACM
SIGCOMM Computer Communication Review. 2018, vol. 48(3), pp. 35–41.
[10] Shannigrahi S., Fan C., Partridge C. ‘What’s in a name? Naming big science
data in named data networking’. Proceedings of the 7th ACM Conference on
Information-­Centric Networking; Canada; 2020. pp. 12–22.
 Chapter 17
Intelligent and cost-­effective mechanism for
monitoring road quality using machine learning
Prabhat Singh 1,2, Abhay Bansal 2, Ahmad E Kamal 3, and
Sunil Kumar 2

Nowadays, one of the most significant components of road infrastructure is moni-

toring road surface conditions, which leads to better driving conditions and reduces
the chance of a road accident. Traditional road condition monitoring systems are
incapable of gathering real-­time information concerning road conditions. In previ-
ous generations, road surface condition monitoring was done for fixed roadways and
vehicles travelling at a constant pace. Several systems have presented a method for
exploiting the sensors installed in automobiles. However, this method will not assist
in forecasting the precise placement of potholes, speed bumps, or staggered roads.
As a result, smartphone-­based road condition evaluation and navigation are
becoming increasingly popular. We propose exploring several machine learn-
ing techniques to accurately assess road conditions using accelerometer, gyro-
scope, and Global Positioning System (GPS) data collected from cellphones.
We also recorded footage of the roadways in order to reduce noise in the data.
This two-­pronged approach to data collection will aid in the exact positioning
of potholes, speed bumps, and staggered roads. This method of data collection
will aid in the classification of road conditions into numerous features such as
roads with smooth surfaces, potholes, speed bumps and staggered highways
using machine learning algorithms. machine learning algorithms are used to
create characteristics such as smooth roads, potholes, speed breakers, and stag-
gered highways. The user will receive this information via the map, which
will classify the various road conditions. Accelerometers and gyroscope sen-
sors will analyze multiple features from all three axes of the sensors in order
to produce a more precise location of designated routes. To classify the road
conditions, we investigate the performance utilizing support vector machine
(SVM), random forest, neural network and deep neural network. As a result, our

1
ABES Engineering College, Ghaziabad, India
2
Amity University, Noida, India
3
IOWA State University, AMES, IA, USA
372  Intelligent network design driven by Big Data analytics

findings demonstrate that models trained using a dual data gathering strategy
will produce more accurate outcomes. Data classification will be substantially
more accurate when neural networks, are used. The methods described here
can be used on a broader scale to monitor roads for problems that pose a safety
concern to commuters and to give maintenance data to appropriate authorities.

17.1 Introduction

Nowadays, roads are the primary resource of infrastructure that provides support
for the migration of human beings, goods and logistics, which in turn will develop
a strong foundation for social and economic development of a country and inter-
connecting city. Apart from this, a sudden increase in volume of vehicles on roads
had caused problems such as traffic accidents, traffic jams, and traffic congestion.
In the last few years, there had been a sudden rise in the urbanization sector. Due
to the increased rates of urbanization, more and more vehicles are being purchased
and made available on the road in order to improve the migration of human beings,
goods and logistics. As the number of vehicles on the road increases, it in turn will
directly increase the traffic congestion, accidents, and pollution.
The Intelligent Traffic and Transportation Management System (TMS) is a
better road and traffic management system that integrates data, provides commu-
nication and provides access technologies to properly integrate and inform drivers,
vehicles and roads in a way that assists people driving the vehicles. TMS’s main
goal is to use information and communication technology to tackle not only car/bus
traffic problems but also economic problems such as ageing, tourism revitalization
and long-­term economic development of a country.

17.1.1 Definition of TMS
TMS is a new transportation system that uses various new technologies to connect
people, roads and vehicles in an information and communication network to solve
various traffic and transportation issues such as traffic accidents and congestion.
In terms of technology, this is a road or traffic information management system in
which road traffic data are gathered and sent to drivers via sensors put along the
roadside. The TMS technique provides users with a variety of road traffic applica-
tions. The components of TMS are represented in (Figure 17.1).

17.1.2 Issues and challenges faced in TMS

Various issues and challenges are being faced in a road TMS, which are as follows:

1. Shifting of population from rural cities to urban cities:

In 21st century, the population in urban cities has increased at a very fast
rate compared to rural areas. In the year 2019, 56% of the total world
 Intelligent and cost-­effective mechanism for monitoring road quality  373

Figure 17.1   Basic components used in TMS

population is living in the urban sector. By the year 2045, UN expects that
76% of the world’s entire population will live in cities or the urban sector.
Assuming these hypotheses to be correct and the rapid increase in urbani-
zation continues, it is also reasonable to expect that when the count of
vehicles increases, other road traffic difficulties occur, such as traffic con-
gestion, traffic accidents and environmental effects. The pictures of traffic
congestions in Asian cities are given in (Figure 17.2).
2. Increased population in urban areas needs transportation services:

As per the rise of population in urban areas, there is a tremendous require-

ment of transportation facility in terms of road, rail and air. It has been
estimated with the rise of passenger traffic by 120–230% by 2045.

3. Increase in demand for transportation service provides sudden increase in CO2:

There is an increase in demand for transport in non-­OECD countries where

passenger traffic is expected to increase by 220–400%. Due to large traf-
fic on roads, CO2 emissions are expected to increase on the passenger by
30–110%.
374  Intelligent network design driven by Big Data analytics

Figure 17.2   Examples of traffic congestions in Asian cities

4. Aggravation of traffic safety:

Traffic safety is one of the major issues nowadays as the requirement and
availability of transport are increasing at a very fast rate. Globally, 1 mil-
lion people have lost their lives and about 50 million got injured due to
traffic accidents. Traffic accidents are one of the primary causes of acci-
dents in younger generations nowadays. In detail, mostly, victims of road
accidents are pedestrians, bicycle drivers and rickshaw drivers.

5. Aggravation of traffic congestion:

There is an increasing number of cars in the urban cities, and almost each
and every individual is having a car, bike or any other means of transport in
urban cities. This rapid increase in turn will give rise to traffic congestion
in mega-­urban cities.

By 2050, 68% of the world’s population will be living in cities, according to

the current estimates. This is putting unprecedented stress on what is, in most cases,
already ageing facilities including transport infrastructure. Noida is an example of
this scenario. It has a growing population, now estimated to be 10 million people,
and still growing at around 8.2% compared with the national average of 2.5%.
 Intelligent and cost-­effective mechanism for monitoring road quality  375

Figure 17.3  Road accidents caused by road conditions in the past 5 years: a

comparative analysis

A key feature of the transport system in Noida is the over-­reliance on road

transport. Rail transport is practically now non-­existent putting further pressure on
the road network. The road network infrastructure is often in a poor and deplor-
able state, inadequate to meet needs, and as often lack of funds for maintenance
and development. These result in a public transport system with unrestricted access
and devoid of standards as well as the enforcement of traffic laws and regulations
leading to almost total chaos and anarchy on the roads. According to the National
Road Safety Commission between January and February of 2018, there were 2,085
crashes with 366 fatalities. Road safety is thus a major problem. In addition, tradi-
tional techniques such as SCOOT loops used in developed countries would not be
very effective because of the volume of transport users involved.
According to WHO statistics, road traffic collisions are now one of the main
reasons for death worldwide. Road accidents, in particular, took about 1.35 million
lives each year (2018). The year wise road accident data is given in (Figure 17.3).
According to studies, the majority of vehicle accidents are caused by poor road
conditions. As a result, a road surface condition monitoring system is required to
increase traffic safety, prevent accidents and safeguard vehicles from damage caused
by poor road conditions. Both road management and drivers are interested in gather-
ing enough details on the quality of the road infrastructure. In consolidated ways,
monitoring road conditions, expensive and sophisticated equipment such as the spe-
cialist accelerometers with data collection systems are employed. These methods
are expensive to build and maintain, and they demand a lot of manual labour, which
can lead to errors while deploying or collecting data. As a result, new solutions to
the country’s transportation difficulties must be found, which are low-­cost, effective,
deployable and scalable.
376  Intelligent network design driven by Big Data analytics

Due to the growth of population as well as city expansion, there is a rise in traf-
fic as well as the volume due to which there is a deterioration of different transporta-
tion infrastructures. Also, the different causes for such destruction are the various
effects of bad construction material use, deficiency present in the design of the road,
climate change. In combination, all these issues basically result in the appearance of
different anomalies in the roads like bumps, potholes, cracks, which can be found in
different expressways highways and streets present worldwide.
Not only these anomalies but also illegal construction of the safety breakers is
also becoming a threat leading to the deterioration of the road safety. Speed breakers
are generally constructed in order to provide safety for the pedestrians in different
zones in order to control the limit of speed for the vehicles, which helps in avoiding
accidents.
Unnecessarily, a large number of unapproved speed breakers have been
installed, many of which are known to not adhere to the actual size set by the NHA.
The speed bumps are very common in the countries like India due to the fact that
different signboards like speed limit, stop, yield and so on will not be able to work
because of a lack of the traffic to enforcement resources.
This generally happens in the country because of various reasons such as higher
speed of the vehicle, negligence of the driver, lesser visibility at night. There are
different incidents that have been reported where vehicles such as cars, scooters,
motorcycles are much more vulnerable because some speed breakers that are unno-
ticed may cause them to lose balance and can lead to severe accidents as well as
damages.
In the year 2016, India had 300,000 road deaths that is actually double the
amount of the estimation by the government that was only 151,000 in accordance
to the 2018 global safety report provided by WHO, which basically highlighted the
lack of the amount of data on fertility road accidents.
According to the data from the government (Figure 17.4), about 150,000 people
have lost their lives in 2017 due to road accidents, which basically means that 17
people die due to road accidents every hour.
The number of deaths accounts for one-­third of the overall accidents, which
averages roughly 53 every hour.
UP has been hit hardest, followed by Tamil Nadu. Delhi is considered to be one
of the safest places for driving considering the various numbers of fatalities that take
place.
Different solutions have been considered in order to automatically detect as well
as report any kind of anomalies on the road to different government agencies so that
the maintenance tasks can be accelerated.
For example, by making the use of different computer vision methods that are
based on texture differences as well as shape segmentation in order to identify the
potholes. Approaches similar to this like contour information, edges, adopting shape
and so on have been considered.
In today’s time, the majority of people possess android mobile phones or smart-
phones that are well equipped with various inbuilt apps such as navigation and Google
maps and sensation systems like Gyro Sensor, Accelerometer, Magnetometer and so
 Intelligent and cost-­effective mechanism for monitoring road quality  377

Figure 17.4   State-­wise road accident analysis

on, are enabled with GPS, and are always connected to the Internet. The technology
of smartphones has been adopted in order to tackle such a problem provided its geo-­
referencing and sensing capabilities.
The accelerometers in smartphones can detect the movement of the device
in such a way that if the vehicle comes across any irregular road surface like a
bomb or a pothole, the accelerometer will record the occurrence of this event. The
major problem then is the identification of the series of continuous readings of the
378  Intelligent network design driven by Big Data analytics

accelerometer when any kind of anomaly occurs. Even though the task of detecting
road imperfections with a smartphone is well defined, the organization has been
unable to gather much more knowledge and a comprehensive perspective due to a
number of issues. It also allows and encourages the development of an Android app
that uses the data collected by these sensors in the user’s smartphone to warn the
user about bumps and poor road conditions.
A literature review of various machine learning algorithms will be discussed in
detail in Section 17.2.

17.2 Literature review of machine learning for road condition

detection

Detection algorithms are the most differing aspect of the previous work on this topic.
Some papers have described machine learning approaches to road quality classifi-
cation; however, others have opted for different solutions. This subchapter aims to
provide an overview of the different feature sets and techniques used in machine
learning approaches (Table 17.1).
The Pothole Patrol is a sensor-­based programme that monitors the condition of
the road surface. It necessitates the integration of particular hardware: an embedded
computer running Linux for data processing, a Wi-­Fi card for data transmission, an
external GPS for localization and a three-­axis accelerometer for road surface moni-
toring are all required for each vehicle. It detects potholes using a machine-­learning
technique. Microsoft developed Nericell, a system that checks traffic conditions. It
necessitates a complex hardware and software configuration. A microphone, GPS
and the SparkfunWiTilt accelerometer are among the external sensors used. The
technology may mistake smooth, uneven and rocky roads due to the inaccuracy of
the detection. Mednis and colleagues proposed a real-­time pothole detection system.
The system makes use of Android phones with accelerometer sensors and basic
algorithms for detecting events from acceleration data. The real positive rate was
found to be 90% in the experiments. The system only uses an accelerometer sensor,
and data are acquired using specialized hardware, which is one of the work’s limita-
tions. A camcorder mounted on the front passenger seat’s headrest is used to label
the items. Labelling driving data with video, on the other hand, is a time-­consuming
and error-­prone task.
The author describes a technique for detecting potholes. The neural network
approach, which has an accuracy of 90–95%, is used to justify the threshold values.
Smartphone accelerometers and gyroscopes are utilized to detect surface irregulari-
ties using an auditory data tagging method in which a labeler sits behind the driver
inside the car and notes everything pertinent he sees or feels. Then, with a 90%
accuracy, SVM is employed for detection and categorization of aberrations.
Using a low-­cost Kinect sensor, Moazzam et al. calculate the volume of a pot-
hole. The use of infrared technology for measurement based on a Kinect sensor is
still a novel concept, and more study is needed to reduce error rates. Zhang et al.
 Intelligent and cost-­effective mechanism for monitoring road quality  379

Table 17.1   Comparison with related research in road surface monitoring

Smartphone External Detection

Reference sensors hardware Data labelling methods
[1] Not used Ultrasonic’s Not mentioned Threshold
Sensors, GPS
[2] Accelerometer, Not used Not mentioned Threshold
Microphone,
GPS
[3] Not used Accelerometer, Manually Threshold/
GPS machine
learning
algorithms
[4] Accelerometer Not used Not mentioned Threshold
[5] Accelerometer, Not used Video SVM
GPS
[6] Accelerometer, Not used Not mentioned Threshold/neural
GPS network
[7] Accelerometer, Not used Audiovisual SVM
Gyroscope
[8] Accelerometer, Not used Audiovisual SVM
Gyroscope
[9] Not used Kinetic sensors Not mentioned Three-­dimensional
(3D)
reconstruction
[10] Not used Stereo camera Not mentioned Image processing
images algorithms
[11] Not used Camera images Manually Three-­dimensional
(3D)
reconstruction
[12] Not used Near-­infrared Manually Video processing
camera methods
[13] GPS, power Not used Manually Threshold on
relative
roughness
[14] Accelerometer, Not used Not mentioned Threshold
GPS
[15] Accelerometer, Not used Not mentioned Threshold, the
GPS standard
deviation of
measurements
Accelerometer Not used Not mentioned Threshold

[16]
Road Sense Accelerometer, Not used Automatic C4.5 Decision tree
Gyroscope,
GPS
380  Intelligent network design driven by Big Data analytics

used stereo camera images combined with a disparity calculation methodology to

find potholes using image processing methods.
Although, in the general field of pothole detection, camera-­based techniques
have proven popular, a NIR camera was utilized in a recent study to classify a range
of road conditions in the lab and in the field.
Video processing approaches, on the other hand, have the disadvantage of being
too computationally expensive for smartphone applications. Vehicles are equipped
with an ultrasonic transducer for ultrasonic techniques.
Ultrasonic waves are continuously transmitted onto road surfaces, and anoma-
lies are discovered by measuring the time it takes for them to return. However,
an expensive measurement apparatus is necessary to produce highly accurate data.
Furthermore, S-­Road Assist uses data received from smartphone sensors to detect
road surface and traffic situations using threshold-­based algorithms.
The standard deviation of readings from mobile-­embedded accelerometers is
typically a threshold in related studies to detect road irregularities.
Seraj and colleagues employed SVMs with RBF kernels for classification. Time
domain, frequency domain and wavelet decomposition were used to extract the
characteristics. The signal’s frequency-­domain representation was created using a
Hamming window function FFT approach, and the signal was decomposed using
a stationary wavelet transform. The time-­domain features included standard devi-
ation, mean, variance, root mean square, peak to peak, mean of absolute values,
zero-­crossing rate, correlation across all axes, tilt angles, signal magnitude area and
waveform duration. Extracted features in the frequency domain included mean fre-
quency, energy of the frequency bands and median frequency. Features extracted
from wavelet decomposition included standard deviation, absolute mean, variance
and energy for every level of detail. It is also mentioned that experimentations were
done with raw signal and demodulated signal data, in order to see which feature sets
provide the best detection results. After feature extraction, a two-­step classifica-
tion process was described. All of the windows were processed in the first stage,
and those with road abnormalities were sent to the second step. Other classifiers
attempted to divide the type of road anomaly present in each window in the second
stage. A sliding window with a 66% overlap was employed in the training data,
which comprised 3,066 windows. The overall accuracy of detecting serious road
irregularities was estimated to be around 90%.
Hoffmann et al. [17] categorize their data using machine learning. Direct road
surface classification and bump detection-­based classification were the two types
of road surface classification methodologies used. A road surface was classified
as smooth, bumpy or rough using the direct road surface classification method.
Features were extracted from GPS and accelerometer data and included: accelera-
tion mean, acceleration variance, and acceleration standard deviation are all vari-
ables to consider. In the direct road classification approach, roads were divided into
sets of segments. For each segment that needs to be classified, previous segments’
features are also taken into account. A feature set optimization was also carried out,
with the purpose of finding those features that increase the classification accuracy
the most. It was found that speed and inclination confused the classifier or didn’t
 Intelligent and cost-­effective mechanism for monitoring road quality  381

contribute to the learning algorithm. For classification, two machine learning algo-
rithms were used: KNN and Naive Bayesian classifier. In order to test the classifiers
on evaluation data, 10-­fold cross-­validation was used. It was noted that while Naive
Bayes and KNN algorithms performed mostly similar, a slight increase was seen
with the KNN. However, the overall accuracy of 78% proved to be not satisfactory.
In the bump detection-­based classification approach, the classifier had to distinguish
between a smooth road and a road that has a bump or some other similar anomaly.
Here, classification accuracy was a lot better, with an accuracy of around 98%.
Eriksson et al. [18] employed signal processing and machine learning to detect
and classify a number of different road irregularities. The topic of other undesired
abnormalities manifesting in the accelerometer signal, such as abrupt turning or
stopping, is discussed in the study. To reject such events, several features, such
as speed and filters, such as a high-­pass filter, were utilized. The training of the
learning algorithm is based on the peak X-­axis and Z-­axis acceleration values and
the instantaneous velocity of the car. To help remove false positives and increase
the overall accuracy of correct anomaly detection, the paper describes a process
where an anomaly is reported only when several other detectors have also detected
an anomaly in the same spot. Overall, the paper concludes that by using training
data, which had been carefully examined, the described pothole detection system
achieved a false positive rate of 0.2% in controlled experiments.
Smartphones and tablets have grown ubiquitous in today’s culture. The world-
wide smartphone market has seen a tremendous increase in shipments in recent years
[19]. Furthermore, in recent years, mobile internet usage has significantly increased,
lately surpassing desktop internet usage on a global scale [20]. Because of the ubiq-
uity of mobile devices and their expanding computational capacity, smartphones are
now capable of completing an increasing number of tasks. Mobile applications on
platforms like Google Android and Apple iOS can make use of the devices’ inherent
sensors, allowing for new ways to use these smart gadgets. Sensors can be used for
a variety of activities, such as detecting the device’s position, or more complicated
difficulties, such as sensing human activity [21].
Another key aspect of modern life is road infrastructure. A vast number of indi-
viduals use roads on a regular basis, whether driving their own car, riding their bikes,
walking, or taking public transportation. Many government services and businesses
rely on the road network in some form. A well-­maintained and safe road network is
beneficial for everyone involved as the rate of motorization continues to rise [22].
Road quality is a reflection of a country’s development status and has an
impact on travel speeds and safety. There is a major need of evaluation of the qual-
ity of roads. The collected data will be forwarded to a self-­built server, where an
analysis module will calculate and analyze the data in order to generate road condi-
tion information. The related previous works revolved around the threshold-­based
detection techniques, but they were not that much effective as they just detected the
damage to roads. Some of the works include the use of fast Fourier transform to get
the acceleration information. This research, on the other hand, makes advantage of
mobile sensor capabilities as well as social community data collecting. Some works
also used spatial–temporal anomalies to detect the damage to the roads. During
382  Intelligent network design driven by Big Data analytics

every action, mobile sensing technology is utilized to identify targeted data uti-
lizing relevant sensing components. This research will use mobile crowd sensing
to gather data and develop collective intelligence. This study employs an unsu-
pervised anomaly detection algorithm to aid in the comprehension of typical and
aberrant road sections. The sensing components of phones are employed in this
paper to capture oscillating amplitude data experienced by a user while driving on
a road. Every second, data are gathered and delivered to a self-­built analytic server.
As a result, in the context of this study, different users traverse the same road and
multiple sets of oscillation amplitude data are obtained for the same road. After
collection, the server uses an aberration finding algorithm to extract the aberrant
information from the data series. The basic goal is to distinguish between typi-
cal and aberrant road portions. This work uses an unsupervised anomaly detection
approach to model road quality and interpret normal and abnormal road sections.
In addition to the analysis of the obtained data and anomaly detection, the Google
Maps SDK is employed.
A road monitoring technique was presented to provide security and ease of use
to various road users [23]. The major goal of this study is to develop a real-­time
Android application called Road Sense that uses a tri-­axial accelerometer and a
gyroscope to automatically forecast road quality. The road position trace will be
displayed on a geographic map utilizing GPS, and all recorded workout entries will
be saved. It will also provide a depiction of a region’s road quality map. The pro-
posed road monitoring system would be created using machine learning, with the
C4.5 Decision tree classifier being used to categorize road segments and build the
model using training data. The overall result shows a consistent accuracy of 98.6%.
Discussion on various road monitoring technologies to predict the road conditions
to provide smooth, safe and comfort travel with less damage to vehicles using 27
papers. The literature survey highlights the hardware and software configurations
and limitations related to developed applications. The authors highlight the various
classification algorithms with the focus to achieve highest accuracy. The framework
of the proposed system is divided into two phases:
1. Training phase: Effective features are extracted from specified sorts of road
conditions depending on acceleration and rotation around gravity first in the Feature
Extraction step. The features are then fed into a classifier model that can do fine-­
grained identification.
2. Prediction phase: It will detect and identify road conditions by sensing real-­
time vehicular dynamics. Pre-­processing is done on sensor readings after collecting
the readings from the accelerometer and gyroscope integrated into the smartphone.
The system would next use the learned classifier model to forecast road quality. In
this paper, the performance of the system is evaluated in two steps: (a) analytical
validation and (b) experimental validation. Analytical validation evaluates the per-
formance of various classifiers using a variety of parameters. The system’s practi-
cality is tested in a real-­world setting during experimental validation. To obtain the
dataset, a drive of about 40 min in length (25 km), and a total of 2,000 samples of
data were collected. The author has compared the performance of three classifiers
named SVM, Navies Bayes and C4.5 on the scale of various parameters such as
 Intelligent and cost-­effective mechanism for monitoring road quality  383

Precision, Recall, ROC area, F-­measure, TP rate and FP rate. Overall, the results
demonstrate that C4.5 is superior in terms of detection accuracy (98.6%).
[24] The focus is on data from traffic accidents, which is the most fundamen-
tal measure of safety without which the scope and nature of road safety cannot be
determined. The precision of the data, data keeping and its analysis are all important
aspects in making successful use of accident information. As a result, the incident
report should be of high quality. If the initial incident report is weak and incom-
plete, analysis and application of the findings are poor as well. Road accidents are
unpredictably unpredictable incidents that require thorough investigation. The het-
erogeneity problem is a fundamental issue in accident data analysis. Although seg-
mentation can be used to reduce data heterogeneity, there is no assurance that it will
produce the most accurate group segmentation, which includes road accidents. The
k-­means technique is used to suggest a model for analyzing road accident data.
Estimating the count of clusters: Estimating the count is the most difficult part of
the clustering method. One of the constraints of this technique is that the value of k
must be provided by the user. If the value of k is incorrect, the clustering results may
be erroneous. Gap statistics are utilized to circumvent this constraint.
Association rule: This mining is the act of generating a set of rules that define
the basic patterns in a data set. The frequency with which attributes of data occur in
the data set determines their correlation. The most fundamental prerequisite of clus-
ter analysis is to figure out how many clusters the clustering method will produce.
After determining the number of clusters, use the R statistical software to partition
the accident data sets using the k-­means clustering algorithm. Cluster-­based acci-
dent variables are determined by a thorough examination of each cluster. Using k to
represent clustering and association rules mining methods, this research proposes a
method for assessing accident patterns of various types of road accidents. The study
looked at accidents that occurred on Maharashtra’s roads in 2015 and 2016 [6-­9].
K-­means clustering finds five categories based on the attribute’s accident type, road
type, light condition and road attributes (C1–C5). Association rule mining was used
to construct rules for each cluster as well as the overall data set. All cluster rules
reveal situations relating to incidents within the cluster.
[25]. In paper [25], SVM is used as classification algorithm. The mobile app
will first process the data via various algorithms and then will transfer the data to
the database by internet. This app produces a sound on pothole detection. With the
help of the data, potholes can be categorized as low, medium and high. The paper
also shows the technologies and calculation methods used in the model in detail and
also shows the settings and results of the experiment. Various research papers are
studied by the authors. One of the papers provides them the parameters like speed,
Z-­peak, etc. which can help in classifying the potholes. One of the research papers
uses accelerometer and GPS same as suggested in this paper by the authors. Various
designs like Cartel System, Pothole Patrol System, Nericell System, etc. can also
be used to detect conditions of road by mobile sensors. The solution is to develop
an Android app that collects data. The app can start or stop or save data collection.
The file in this paper is saved as .csv format. With the help of accelerometer sen-
sors readings are captured. along with these readings. Accelerometer readings are
384  Intelligent network design driven by Big Data analytics

also captured. The changes/deviation in accelerometer readings detects the potholes.

K-­means clustering is used for depicting the pattern. This algorithm clusters data
into groups: potholes and non-­potholes. After the readings are recorded, script file
is then allowed to run on the saved file provided by the android app and labeling is
done. Then ML models are used to train the data. Before training, we apply cluster-
ing for segregation of pothole and non-­pothole. In this paper, accuracy is calculated
with various classifiers and SVM proved to provide maximum accuracy of 99.6%.
[26The paper [26] describes that road networks connecting different residences,
villages, cities and even countries are becoming key components of contemporary
transportation infrastructure. . In the real world, detecting road conditions, which
may be separated into four stages, is critical. These are as follows:

• Manual detection. Manual measurements have a number of drawbacks, includ-

ing a high cost, a long time to complete and a low efficiency.
• Semi-­automated detection. Computers control the mechanical activity, save
data in real time and analyze and calculate it at predefined intervals.
• Automated detection. Ultrasonic, laser, ground-­penetrating radar and other
technologies are used to identify roads.
• Informatization detection. Communication, network, database and other
information technologies are used in road quality testing. Highly developed
and shared information technology and information resources can significantly
improve test accuracy in any discipline.

A taxi-­based mobile sensing system named Pothole Patrol was shown for moni-
toring and analyzing pavement conditions. Three-­axis acceleration sensors, a GPS
receiver and a laptop computer with a data aggregation and processing unit are all
included in the cab. A more efficient and cost-­effective technology for identifying
potholes and cracks was devised with laser imaging. It used sensors implanted in
cell phones to collect data in cars. The system, for instance, may employ a one-­
degree-­of-­freedom vibration model to actively gain a vehicle’s competence. Both
supervised and unsupervised machine learning methods are used to detect road con-
ditions. Using a simple machine learning methodology and a clustering algorithm, a
method for detecting road anomalies was presented for evaluating driving behavior.
In the case of potholes and bumps, the average detection rates were determined to be
88.66 and 88.89%, respectively. The crowdsourcing strategy not only lowers costs
and improves efficiency but also removes geographical and time constraints, making
communication more convenient. The proposed solution of the paper is as follows:
A. System Model: Dr. Taguchi employed the Mahalanobis distance (MD) and
measurement scale and provided a threshold to identify unknown samples using
a combination of MD and signal-­to-­noise ratio.
B. Benchmark space: The window size of flat road data is initially believed to be
“h”. The flat road sample space has m dimensions, and the sample data is Xi =
(Xi1, Xi2, Xim), with the i-­th feature’s data acquired in the j-­th time.
 Intelligent and cost-­effective mechanism for monitoring road quality  385

C. Validation of benchmark Space: We need to collect some unusual road data to

test the validity of the benchmark space Yi = (Yi1, Yi2, · · · , Yim).
D. Optimization of spatial Features: In general, the characteristic variables
include duplicate information, resulting in a miscalculation and a negative influ-
ence on classification outcomes.
Our study took place in Xi’an, where we collected 34,895 records of vehicle
traffic data over a period of 15 weeks. We design and then standardize a benchmark
space using the detection model validation by computing the correlation matrix and
the MD of space by 30 eigenvectors of flat road signal.

17.3 Gaps identified in the literature

After doing the literature survey in detail, the following are the gaps identified in the
mentioned research domain:

a. SCOOT Technology can only be used where the traffic flow is very less such as
Yamuna Express Way or Agra-­Luck now Expressway.
b. In addition to SCOOT, a new technology was developed that is ETSI-­G5, which
is very expensive as it uses onboard units (OBUs) and roadside units (RSUs)
that are placed along with the road infrastructure.
c. Neither SCOOT nor ETSI-­G5 talks about the safety of roads. Only monitoring
or surveillance of vehicles is done using these technologies. This seems to be
providing a costly incomplete solution in terms of all types of traffic flow.
d. In the literature, various road safety techniques have been suggested; however,
in maximum cases focus was on the driver’s fault for the road accidents.
e. Quality of the roads and accidents due to animals is not being considered in the
literature as the factor for road accidents or damages to the vehicles.
f. Study of traffic congestion at traffic lights was not discussed in the literature.

17.4 Proposed methodology

The challenge is to build a low-­cost information platform for monitoring traffic

flows and to bring that information from different parts of a city together so that it
can be analyzed to enable better traffic management routines for mega-­cities in low-­
income countries. The present traffic flow monitoring techniques used in developed
countries such as SCOOT cannot cope with the volume of traffic, which would go
quickly wear out the SCOOT loops. In addition, newer systems such as ETSI-­G5
technology for connected vehicles are very expensive. These systems work by hav-
ing OBUs and RSUs, which are placed along the road infrastructure. Currently,
an OBU and an RSU each cost over US$1,000 . So, the solution being proposed
is to use Bluetooth Low Power (BLE) devices. These devices communicate using
Bluetooth technology and have a common coverage range of 70–100 m.
386  Intelligent network design driven by Big Data analytics

Since mobile phones also have Bluetooth interfaces, a lot of apps have been
written to take data from the BLE device and use the mobile phone to communicate
with a central server or navigational entity. Though what we are proposing does not
remove such capability, the system we want to build is not dependent on users hav-
ing mobile phones. Instead, the BLE device will beacon every few seconds and those
beacons with their UUIDs will be picked up by Wi-­Fi/BLE gateways. The gateway
devices cost £54 each, and this project will buy 50 such gateways. These gateways
will then use eight wireless routers, which will be connected to the University of
Ghana’s communications network. This is an Ethernet network running at 1 Gbps.
The information will be sent to a central server, where it will be stored, processed
and used to display real-­time traffic flows.
The Wi-­Fi/BLE gateways will be placed on two very busy roads that border
the University of Ghana. The first is called National Highway – four or N4, which
runs from north to south along the east side of the Campus. In fact, it forms a direct
boundary with the campus for 3.5 km. The second road is called the Haatso-­Atomic
road and runs from east to west on the north side of the campus and forms a natural
campus boundary for 1.5 km. So, in this proposal, we would like to monitor traffic
from both roads making this a total of 5 km of road infrastructure that will be moni-
tored. The data gathered will be analyzed to show the traffic flow along these roads.
The results will be displayed using a Virtual Network Computing server; hence,
the results will be available to all commuters in real time. Information will also be
communicated using mobile phones as well as road signage such as overhead road
gantries.
Research methodology has been organized into:

1. Research design
The research is exploratory as well as descriptive. The study’s associated
variables were discovered, and the research problem was specified using an
exploratory research methodology. Exploratory research also proposes a con-
ceptual framework that incorporates the pertinent variables. Following that, a
descriptive research methodology was employed to empirically test the stated
conceptual framework of the study, as well as statistical analysis of the data.
2. Sampling design
The sample for the study of the city of Noida has been selected. We plan to
collect the data regarding the roads condition and traffic conditions of the road
near the Amity campus. To study will be performed on the road conditions
using the sample size of 500 vehicles with BLE devices for communication and
sensors like accelerometers embedded with them.
3. Data collection instruments
The basic data collection instruments are the sensor that is embedded in the
vehicle.
1. Accelerometer
2. Gyro meter
3. Cameras
4. GPS
 Intelligent and cost-­effective mechanism for monitoring road quality  387

5. Smartphones with Wi-­Fi connectivity

6. Bluetooth
7. APP on smartphone for emergency messages
4. Data collection method
The study is based on the survey method of collecting initial information
during the research process. Primary data are collected by the researcher him-
self using the survey method. It is a reliable way to collect the data because the
researcher knows where it came from and how it was collected and analyzed.
5. Statistical techniques
The various data cleaning and analysis operations will be performed based
on the data collected from the sensors and perform the analysis using statically
operations used in the machine learning algorithm. The summarized tentative
plan for the same has been described in Figure 17.5.

17.5 Implementation

17.5.1   Flutter-based application

We created a flutter-­based android app to collect data. This software uses smartphone
sensors to collect sensor data (accelerometer and gyroscope) for potholes and speed
bumps (Figure 17.6) . We put a mobile device with an application on the dashboard of
the vehicle. The programme puts the obtained data in a Firebase database for machine
learning model training when the data gathering operation is done. Regarding the
data gathering phase, a machine learning algorithm is trained on the collected dataset
to recognize distinct anomalies (potholes, not potholes and speed bumps), and the
present position is projected as a pothole, not pothole, or speed bump (Figure 17.7).

Figure 17.5   Block diagram of the proposed methodology

388  Intelligent network design driven by Big Data analytics

Figure 17.6   Accelerometer and gyroscope sensor reading on a mobile device

17.5.2 Machine learning models

Before training any machine learning algorithm, the gathered data must be pre-­
processed. To reduce noise from data, which if not eliminated can have an influence
on our machine learning model’s eventual outputs. The machine learning model will
be better at generating predictions if the data we give it is of greater quality. So, in
order to execute data pre-­processing and data cleaning, we first performed various
pre-­processing tasks, such as translating timestamp strings into the proper date and
time format. We developed statistical characteristics of accelerometer and gyroscope
readings for feature extraction and feature engineering, such as mean, min, max and
standard deviation for all axes (x, y and z). We standardized data features for algo-
rithms that are sensitive to varying data scales. The volume of the data has an impact
on algorithms that use distance-­based metrics as part of their implementation.
As can be seen in Figure 17.8, the data we collected were extremely skewed.
The dataset’s imbalance should be carefully managed since it might have a negative
impact on the end outcomes as well as the machine learning models’ ultimate inter-
pretation. Dealing with the dataset’s imbalance can be done in a number of ways.
It is possible that it is not gathering enough data for each class, oversampling the
minority class (one in the example above) and undersampling the majority class (0
 Intelligent and cost-­effective mechanism for monitoring road quality  389

Figure 17.7  (a) and (b) Represent the working model for identifying whether
there is no pothole or not a pothole on the road, respectively.

represents the majority class in the above case). We dealt with the problem by adopt-
ing stratified cross-­validation processes. The entire dataset is divided into numerous
folds, with the target distribution remaining the same for each fold. It is called strati-
fied because the folds are made using stratified sampling.
As our initial model, we used logistic regression. The logistic function, com-
monly known as the sigmoid function, is used to create the simplest basic categori-
zation approach. It is an S-­shaped curve that transforms all real numbers to a number
between 0 and 1.
The logistic regression model predicts probabilities. It is given by:
‍Y = 1/(1 + e(x)).‍
The sample logistic regression equation can be modelled as
‍ Y = 1/(1 + e((b0 + b1 x))).‍
The logistic regression equation’s coefficients can be estimated via maximum-­
likelihood estimation. The second model we built was the k closest neighbours
model. KNN is an easy-­to-­use and develop machine learning algorithm. It is a
supervised machine learning technique that may be applied to classification and
390  Intelligent network design driven by Big Data analytics

Figure 17.8   Class distribution of Potholes present (1)/absent (0)

regression issues. It is founded on the idea that similar objects are more likely to
be located near together, i.e., similar items are kept together as much as possible
[5]. The KNN method calculates closeness using several distance metrics such
as Minkowski distance, Euclidean distance and Manhattan distance. There is one
hyperparameter k in this method that we must choose well in order for our algorithm
to perform well on the data. To pick the optimal value of k, we looped over various
neighbours in the range of 5–25 in steps of 5, trained the algorithm and then stored
the best-­performing model. The final model is SVMs. The SVM is a type of super-
vised machine learning approach for solving classification and regression issues. In
an SVM, each data point is displayed in n-­dimensional space, where n is the number
of features in the data. We utilize a hyperplane to easily distinguish the classes while
categorizing them.
The decision tree classifier is the fourth model. A decision tree is a type of
supervised machine learning approach that may be used to solve classification and
regression problems. A decision tree is a hierarchical structure consisting of nodes
and their connections. There are three types of nodes in a decision tree: root node,
internal node, and leaf node. A root node is one that has zero or more outward edges
and no incoming edges. A node with exactly one incoming edge and two or more
outgoing edges is called an internal node. A leaf node has only one inbound edge
and no outward edges. Test attributes are used to analyze and segregate data based
on the characteristics of each non-­leaf node in a decision tree.
Random forest classifier is the fifth model. Random forest is a supervised
machine learning approach for classifying and predicting data. A random forest is
a collection of several decision trees that, as the name suggests, create a forest. It
uses the ensemble approach. Each decision tree forecasts a distinct class label, and
the model’s final forecast is determined by the class label that receives the most
votes. A random forest model outperforms a decision tree model in the majority of
 Intelligent and cost-­effective mechanism for monitoring road quality  391

situations due to the low correlation between many decision trees in the model. The
model’s final prediction is derived using the ensembling approach, which combines
many low-­correlated decision trees to get a range of forecasts. Uncorrelated models
can make more precise forecasts than individual model projections. The reason for
this is that in a random forest, each uncorrelated decision tree stops the others from
making their own mistakes.
Boosting techniques were also utilized to develop numerous models, including
the gradient boosting classifier. It makes use of a boosting method. Boosting is a
method of transforming weak pupils into powerful ones. In this situation, the deci-
sion tree is a poor learner. Algorithms for boosting exist in a range of shapes and
sizes. These algorithms differ from one another in terms of how they transform weak
learners into strong learners or how they uncover and correct weak learners’ flaws
in order to develop strong learners who can more accurately predict data. Adaboost
is a boosting method that addresses a poor learner’s shortcomings by using adaptive
weights or high weight data points. Adaboost starts by creating a decision tree for
each observation with the same weights. In succeeding rounds, it adjusts the given
weights based on the predictability of a single observation. The data that are more
difficult to categorize are given a lower weight, while the data that are simpler to
classify are given a higher weight. A gradient boosting machine, on the other hand,
addresses the drawbacks of a weak learner by using gradients in the loss function.
It uses a progressive, cumulative, and sequential approach to train models. Another
algorithm that uses the boosting technique is Xgboost. If the data isn’t too tough,
it can build an ensemble of linear models and an ensemble of gbtree, which uses a
decision tree as the basis estimate. It works by building a base model that predicts
the target variable first and then training subsequent models to match the residuals
from prior phases. To decrease residuals from previous phases, Xgboost employs a
step-­by-­step strategy.

17.6 Results
As a consequence of our research, we identified a few surprising and fascinating data,
indicating that AI approaches may be used to handle the issue of pothole recognition
and street condition order. We used a number of AI computations and attempted a
variety of proving and boundary modifying processes that we were familiar with to
get to our conclusions (Figure 17.9 a, b, c, d).
We trained several other models, such as K nearest neighbours, SVMs, and
gradient boosting algorithm, in a similar fashion, and they yielded 91.4, 92, and
92.3% accuracy, respectively. Figure 17.10a,b,c illustrate the results of a compara-
tive examination of models based on F1 Score, precision and recall.

17.6.1 Comparison with the existing models

We completed extensive research on this issue statement of pothole detecting.
From data collection through data preprocessing, model training, hyperparameter
modification, model selection and monitoring, various intriguing discoveries were
392  Intelligent network design driven by Big Data analytics

Figure 17.9  (a) Logistic regression, (b) random forest, (c) XG Boost, (d) KNN,

(e) SVM
 Intelligent and cost-­effective mechanism for monitoring road quality  393

Figure 17.10  (a) A comparison of models based on the F1 score, (b) precision-­

based comparative analysis of models, (c) recall-­based
comparative analysis of models
394  Intelligent network design driven by Big Data analytics

discovered as the research progressed. Also, we gathered inspiration from all of the
references and used it to replicate their ideas, which we then improved upon to cre-
ate our own.
A major modification we made as part of our approach was a change in data
gathering that we thought would help us obtain a wider range of results from our
study. As a result, we used a car rather than a motorcycle, which had previously
been employed extensively in other investigations. This technology may be a little
more expensive than previous research, but it allows us to obtain more consistent
and accurate data from mobile sensors. As a result, our effort in this study meets our
expectations in terms of providing us with a high-­quality dataset.
In addition, most researchers solely use accelerometers as their major source
of research, as evidenced by previous studies. However, in order to offer additional
context to the readings supplied by the sensor data, we analyzed the gyroscope read-
ing in addition to the accelerometer reading in this study. We were able to syn-
chronise our accelerometer readings to a global frame of reference thanks to the
gyroscope data, regardless of the car’s orientation or any other disturbances at the
time of data collection.
Last but not least, for our research, we used all three axes of the accelerometer.
This indicates that, unlike most previous studies.

17.7 Conclusion and future scope

As a consequence of our study, we produced an android application using flutter.

It is used to obtain information from mobile sensors (accelerometer and gyroscope
readings). These are speed bumps, not potholes, according to the measurements.
Data on numerous roadways were compiled using the programme. After the data
have been collected, supervised machine learning techniques are used to process
it. Several traditional categorization approaches were applied to the collected data.
Traditional classification approaches included SVM, decision tree, and random for-
est, while modern boosting algorithms included XGboost and CAT boost. There are
several problems in our current work that can be solved in future initiatives. The
data gathered are significantly skewed. You will be able to finish the task completely
if you collect enough samples for each class. The data collection strategy has room
for improvement. We just utilized a single Android smartphone to collect data. This
can be aided by the use of many mobile devices. Crowd-­sensing techniques, on the
other hand, can be used to collect and extract road quality data from a large number
of users, which is then fed into a parent application for more exact findings.

References

[1] Pan J., Khan M., Popa S.I, et al. ‘Proactive vehicle re-­routing strategies for con-
gestion avoidance’. 2012 IEEE 8th International Conference on Distributed
Computing in Sensor Systems (DCOSS), Publisher: IEEE, Hangzhou, China,
 Intelligent and cost-­effective mechanism for monitoring road quality  395

pp. 265-­272, doi: 10.1109/DCOSS.2012.29; New York: IEEE, 2012. pp.

265–72.
[2] Brennand C.A., de Souza A.M., Maia G., et  al. ‘An intelligent transporta-
tion system for detection and control of congested roads in urban centers’.
20th IEEE Symposium on Computers and Communication (ISCC), Larnaca;
NewYork: IEEE, 6–9 July; 2015. pp. 476–81.
[3] De Souza A.M., Yokoyama R.S., Maia G., et  al. ‘Real-­time path planning
to prevent traffic jam through an intelligent transportation system’. IEEE
Symposium on Computers and Communication (ISCC), Messina; NewYork:
IEEE, 27–30 June; 2016. pp. 726–31.
[4] De Souza A.M., Yokoyama R.S., Botega L.C., et al. ‘Scorpion: a solution
using cooperative rerouting to prevent congestion and improve traffic con-
dition’. 2015 IEEE International Conference on Computer and Information
Technology, Liverpool; New York: IEEE, 26–28 October; 2015. pp.
497–503.
[5] Doolan R., Muntean G.M. ‘EcoTrec – a novel VANET-­based approach to
reducing vehicle emissions’. IEEE Transactions on Intelligent Transportation
Systems. 2016, vol. 99, pp. 1–13.
[6] Treiber M., Kesting A., Wilson R.E. ‘Reconstructing the traffic state by fu-
sion of heterogeneous data’. Computer-­Aided Civil and Infrastructure
Engineering. 2011, vol. 26(6), pp. 408–19.
[7] Rehborn H., Koller M., Kerner B.: ‘Traffic datafusion of vehicle data to de-
tect spatiotemporal congested patterns’. 19th ITS World Congress, Vienna,
Austria; Washington: ITS America, 2012. pp. 1–8.
[8] Chen N., Xu X. ‘Information-­fusion method for urban traffic flow based on
evidence theory combining with fuzzy rough set’. Journal of Theoretical and
Applied Information Technology. 2013, vol. 49(2), pp. 560–6.
[9] Pan J., Pan J., Borcea C P.S. ‘Divert: a distributed vehicular traffic re-­routing
system for congestion avoidance’. IEEET Mobile Computer. 2016.
[10] Wang S., Djahel S., McManis J. ‘An adaptive and vanets-­based next road re-­
routing system for unexpected urbantraffic congestion avoidance’. 2015 IEEE
Vehicular Networking Conference (VNC), Columbus, OH; New York: IEEE,
16–18 December; 2015. pp. 196–203.
[11] Madli R., Hebbar S., Pattar P., Golla V. ‘Automatic detection and notification
of potholes and humps on roads to aid drivers’. IEEE Sensors Journal. 2015,
vol. 15(8), pp. 4313–18.
[12] Mohan P., Padmanabhan V.N., Ramjee R. ‘Nericell: rich monitoring of road
and traffic conditions using mobile smartphones’. Proceedings of 6th ACM
Conference Embedded Networks Sensor Systems; 2011. pp. 323–36.
[13] Eriksson J., Girod L., Hull B., Newton R., Madden S., Balakrishnan H. ‘The
pothole patrol: using a mobile sensor network for road surface monitoring’.
Proceedings of 6th International Conference on Mobile Systems, Applications,
and Services; 2008. pp. 29–39.
[14] Mednis A., Strazdins G., Zviedris R., Kanonirs G., Selavo L. ‘Real time pot-
hole detection using Android smartphones with accelerometers’. Proceedings
396  Intelligent network design driven by Big Data analytics

of International Conference on Distributed Computing in Sensor Systems

Workshops; 2011. pp. 1–6.
[15] Perttunen M. ‘Distributed road surface condition monitoring using mobile
phones’. Proceedings of International Conference on Ubiquitous Intelligence
and Computing; 2011. pp. 64–78.
[16] Kulkarni A., Mhalgi N., Gurnani S., Giri N. ‘Pothole detection system using
machine learning on android’. International Journal of Emerging Technology
and Advanced Engineering. 2014, vol. 4(7), pp. 360–4.
[17] Android Developer page for SensorEvent class. Available from https://devel-
oper.android.com/reference/ android/hardware/SensorEvent.html [Accessed
10 May 2017].
[18] Karagiannis G., Altintas O., Ekici E., et al. ‘Vehicular networking: a survey
and tutorial on requirements, architectures, challenges, standards and solu-
tions’. IEEE Communications Surveys & Tutorials. 2011, vol. 13(4), pp.
584–616.
[19] Bajpai P. The evolution of smartphone markets: where growth is going. 2016.
Available from www.nasdaq.com/article/the-evolution-of-smartphone-mar-
kets-where-growth-is-going-cm619105 [Accessed 10 May 2017].
[20] StatCounter press release. Mobile and tablet internet usage exceeds desktop
for first time worldwide. 2016. Available from http://gs.statcounter.com/press/​
mobile-and-tablet-internet-usage-exceeds-desktop-for-first-time-worldwide
[Accessed 10 May 2017].
[21] Anguita D., Ghio A., Oneto L., Parra X., Reyes-­Ortiz J.L. ‘Human activity
recognition on smartphones using a multiclass hardware-­friendly support vec-
tor machine’. International Workshop on Ambient Assisted Living. Springer
Berlin Heidelberg; 2012. pp. 216–23.
[22] Tonde V.P., Jadhav A., Shinde S., Dhoka A., Bablade S. ‘Road quality and
Ghats complexity analysis using android sensors’. International Journal of
Advanced Research in Computer and Communication Engineering. 2015,
vol. 4(3), pp. 101–4.
[23] Hu S., Su L., Liu H., Wang H., Abdelzaher T.F. 'SmartRoad: smartphone-­
based crowd sensing for traffic regulator detection and identification'. ACM
Trans. Sensor Netw. 11, 4, Article 55 (July 2015), 27 Pages. DOI: Http://
Dx.DOI.Org/10.1145/2770876. 2015, vol. 11(4).
[24] Sharma H., Naik S., Jain A., Raman R.K., Reddy R.K., Shet R.B. ‘S-­road
assist: road surface conditions and driving behavior analysis using smart-
phones’. Proceedings of International Conference on Connected Vehicles and
Expo; 2015. pp. 291–6.
[25] Kulkarni K., Prashant K.V., Nasikkar S., Ahuja T., Mhetre N. ‘Predicting road
anomalies using sensors in smartphones’. Imperial Journal of Interdisciplinary
Research. 2016, vol. 2(6), pp. 219–22.
[26] Li Y., Xue F., Feng L., Qu Z. ‘A driving behavior detection system based
on a smartphone’s built-­in sensor’. International Journal of Communication
Systems. 2017, vol. 30(8), p. e3178.
 Chapter 18
Conclusion
Sunil Kumar 1, Glenford Mapp 2, and Korhan Cengiz 3

18.1 Conclusion

This book presented applications, technologies, challenges, and implementation

design models of intelligent networks design by Big Data, Internet of Things, artifi-
cial intelligence (AI), and cloud computing approaches in different sectors. Research
into real-­time fault tolerance, security, and data analytics by addressing significant
data volume and velocity measurement are essential in intelligent networks, for
example, where different analytical models and AI and machine learning algorithms
improve end-­to-­end performance, efficiency, and quality of service (QoS) [1–10].
The availability of Big Data, AI, cloud computing, low-­cost commodity hard-
ware, and new information management and analytic software has produced a
unique moment in the history of intelligent networks [11–21]. The convergence of
these trends means that we have the capabilities required to analyze unique data
sets quickly and cost-­effectively. They represent a genuine leap forward and a clear
opportunity to realize enormous efficiency, productivity, revenue, and profitability.
This book also discussed issues and challenges that can be addressed and over-
come in the future using the new upcoming technologies. The age of intelligent
networks is here, and these are truly revolutionary times if business and technology
professionals continue to work together and deliver on the promise.
Thank you for taking the time to read this book, and we hope you enjoyed read-
ing it as much as we did write it.

References

[1] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘EMEEDP: enhanced
multi-­hop energy efficient distributed protocol for heterogeneous wireless
sensor network’. 2015 5th International Conference on Communication

1
Department of Computer Science & Engineering, Amity University, Uttar Pradesh, India
2
Department of Computer Science, Middlesex University, UK
3
Department of Electrical-­Electronics Engineering, Trakya University, 22030 Edirne, Turkey
398  Intelligent network design driven by Big Data analytics

Systems and Network Technologies, CSNT; Gwalior, India, 04-­06 Apr; 2015.
pp. 194–200.
[2] Kumar S., Ranjan P., Ramaswami R. ‘Energy optimization in distrib-
uted localized wireless sensor networks’. Proceedings of the International
Conference on Issues and Challenges Intelligent Computing Technique
(ICICT); Ghaziabad, India, 07-­08 Feb; 2014.
[3] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, Internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India, 22-­23 Oct; 2021. pp. 01–6.
[4] Sudhakaran S., Kumar S., Ranjan P., Tripathy M.R. ‘Blockchain-­based
transparent and secure decentralized algorithm’. International Conference
on Intelligent Computing and Smart Communication 2019. Algorithms for
Intelligent Systems. Springer; Singapore; 2020.
[5] Kumar S., Trivedi M.C., Ranjan P., Punhani A. Evolution of Software-­Defined
Networking Foundations for IoT and 5G Mobile Networks. 10. IGI Publisher;
2020. p. 350.
[6] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy efficient mul-
tichannel MAC protocol for high traffic applications in heterogeneous wireless
sensor networks’. Recent Advances in Electrical & Electronic Engineering.
2017, vol. 10(3), pp. 223–32.
[7] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘Resource efficient clus-
tering and next hop knowledge based routing in multiple heterogeneous wire-
less sensor networks’. International Journal of Grid and High Performance
Computing. 2017, vol. 9(2), pp. 1–20.
[8] Kumar S., Cengiz K., Vimal S., Suresh A. 'Energy efficient resource mi-
gration based load balance mechanism for high traffic applications iot'.
Wireless Personal Communications. 2021, vol. 10(3), pp. 1–14. doi: 10.1007/
S11277-­021-­08269-­7.
[9] Haidar M., Kumar S,. ‘Smart healthcare system for biomedical and health
care applications using aadhaar and blockchain’. 2021 5th International
Conference on Information Systems and Computer Networks, ISCON 2021;
GLA Mathura, 22-­23 October 2021; 2022. pp. 1–5.
[10] Punhani A., Faujdar N., Kumar S. ‘Design and evaluation of cubic Torus
Network-­on-­Chip architecture’. International Journal of Innovative
Technology and Exploring Engineering (IJITEE). 2019, vol. 8(6), pp.
2278–3075.
[11] Dubey G., Kumar S., Kumar S., Navaney P. ‘Extended opinion lexicon and
ML-­based sentiment analysis of tweets: a novel approach towards accurate
classifier’. International Journal of Computational Vision and Robotics.
2020, vol. 10(6), pp. 505–21.
[12] Singh P., Bansal A., Kamal A.E., Kumar S. ‘Road surface quality monitoring us-
ing machine learning algorithm’ in Reddy A.N.R., Marla D., Favorskaya M.N.,
Satapathy S.C. (eds.). Intelligent Manufacturing and Energy Sustainability.
Smart Innovation, Systems and Technologies. 265. Singapore: Springer; 2022.
 Conclusion  399

[13] Kumar S., Ranjan P., Radhakrishnan R., Tripathy M.R. ‘Energy aware dis-
tributed protocol for heterogeneous wireless sensor network’. International
Journal of Control and Automation. 2015, vol. 8(10), pp. 421–30.
[14] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘A utility maximiza-
tion approach to MAC layer channel access and forwarding’. Progress in
Electromagnetics Research Symposium. 2015, pp. 2363–7.
[15] Kumar S., Ranjan P., Ramaswami R., Tripathy M.R. ‘An NS3 implementa-
tion of physical layer based on 802.11 for utility maximization of WSN’. 2015
International Conference on Computational Intelligence and Communication
Networks; Jabalpur, India, 12-­14 Dec; 2016. pp. 79–84.
[16] Sharma A., Awasthi Y., Kumar S. ‘The role of blockchain, AI and IoT for smart
road traffic management system’. 2020 IEEE India Council International
Subsections Conference, (INDISCON); Visakhapatnam, India, 03-­04 Oct;
2020. pp. 289–96.
[17] Singh P., Bansal A., Kumar S. ‘Performance analysis of various information
platforms for recognizing the quality of indian roads’. Proceedings of the
Confluence 2020 - 10th International Conference on Cloud Computing, Data
Science and Engineering; Noida, India, 2020. pp. 63–76.
[18] Kumar S., Ranjan P., Singh P., Tripathy M.R. ‘Design and implementation of
fault tolerance technique for Internet of things (IoT)’. 2020 12th International
Conference on Computational Intelligence and Communication Networks
(CICN); Bhimtal, India, 25-­26 Sep; 2020. pp. 154–9.
[19] Singh P., Bansal A., Kumar S. ‘Road monitoring and sensing system using
mobile sensors’. 2022 12th International Conference on Cloud Computing,
Data Science & Engineering (Confluence); Noida, India, IEEE, 2022. pp.
165–70.
[20] Reghu S., Kumar S. ‘Development of robust infrastructure in networking to
survive a disaster’. 2019 4th International Conference on Information Systems
and Computer Networks (ISCON ); Mathura, India, 21-­22 Nov; 2019. pp.
250–5.
[21] Chauhan R., Kumar S. ‘Packet loss prediction using artificial intelligence uni-
fied with big data analytics, Internet of things and cloud computing technolo-
gies’. 5th International Conference on Information Systems and Computer
Networks (ISCON); Mathura, India, 22-­23 Oct; 2021. pp. 1–6.
This page intentionally left blank
 Index

actuators 14, 15, 24, 258, 270 association rule mining 61, 63, 67, 383
Ad hoc On-Demand Distance Vector algorithm 67
(AODV) 238–9 ATMEL 89s52 microprocessor 311
Ad hoc On-Demand Distance Vector RP attack detection 361–2
(AODV RP) 238 audio forensics 107
Ad hoc OnDemand Multipath Distance autocorrelation 200
Vector (AOMDV) 238–9
Ad hoc OnDemand Multipath Distance
Vector RP (AOMDV RP) 239 backup 119–20
Adafruit 25 Beaglebone 25
Advance Encryption Standard (AES) biclique attack 172
170–2 Big Data Analytics 15, 96
aim-based networking 26 advantages 100
AMBA Advanced Extensible Interface audio forensics 107
Protocol 40 challenges 16
AMC 236–8, 245, 247 classification 97–9
analog turbidity sensor 216 cloud and computing 16–17
anomaly detection 22 cyber crime 105–9
Application programming interfaces decision trees 107
(APIs) 19, 140–1 distributed processing 17–18
Application-specific integrated circuits in digital forensics 108–6
(ASICs) 2, 35, 42, 46 edge computing 16, 17
Arduino 25, 310, 315, 316, 318, examples 94–5
320, 332 image forensics 107
Arduino IDE 215, 218–220, 316, 320 implementation layer 101
Arduino UNO 309–312, 316, 345 issues 16
Arduino-based health-care life cycle 96–7
application 304 MapReduce 106–7
ARM processors 40 neural networks 107
Artificial intelligence (AI) processing techniques
convolutional neural network (CNN) MapReduce 102–5
with 296 traditional method 101–2
in networks 21, 26 random forests 107
Artificial neural network (ANN) 299, real-time edge analytics for 109–11
341, 342, 349 security issues associated with 108
402  Intelligent network design driven by Big Data analytics

stages of 97 in infrastructure as a service model

working 99–100 (IaaS) model 164–5
Blockly 23–4, 26 in platform as a service model
blood pressure sensor 312 (PaaS) model 163–4
Bluetooth Low Power (BLE) devices in private cloud 165–6
385–7 in public cloud 166
body position sensor (ADXL 335) 313 in software as a service (SaaS)
boosting 391 model 162, 163
broad network access 161 sole features 116
Cloud Disaster Recovery (CDR) 128–9
Cloud Service Providers (CSPs) 160
CAE: See Correlation attribute cloud storage as a service 125
evaluator (CAE) Cluster Based Association Rule Mining
chip verification 39 Grounded on Sequence Number
Cisco Packet Tracer 23 (CBARBSN) 341
classification algorithms 66 clustering 61, 63, 66, 383
Classification and Regression Tree clustering algorithms 66–7
algorithm (CART) algorithm CNN-based finger-vein extraction
72, 75 297–8
cloud-based computing models 109 CNNs: See convolutional neural
cloud computing 16–17, 115–16 networks (CNNs)
advantages 116 CNN-UDRP algorithm 342
characteristics 161–2 Cold/Hot backup strategy 122
data recovery and backup 118–20 Comma-separated value (CSV) files 19
cost-efficient remote backup community cloud 260–1
services for enterprise cloud security issues in 166
127–8 computing cheating attack model 177
multi-cloud-based disaster controllers 14–15
recovery service 128–31 Convolutional neural network (CNN)
for ransomware attack defense 284–6, 288–9
123–4 with AI 296
real-time 120–2 classification 295
and security 122–3 image classifier 297–8
techniques 130 image preprocessing 295
for video compression 124–5 interstitial lung disease detection
disadvantages 117–18 293–5
in healthcare 258–61 layers 296
origin 116 training 295
and security 159–61 training phase 292
security issues training strategies 292
in cloud deployment models 165–6 convolution layers 289–98
in cloud environment 166–9 Correlation attribute evaluator (CAE) 3,
in cloud service model 162–16 61, 70, 76, 80–1, 85, 86
in community cloud 166 CO2 sensor (MQ 135) 314
in hybrid cloud 166 CO sensor (MQ 9) 313
 Index  403

Coverage-Driven Verification (CDV) 42 algorithm 72, 74–5

credit card fraud detection 22 classification algorithm 78, 79, 83,
cryptographic server (CS) 174–176 84, 86
cryptography classifier 382, 390
Advance Encryption Standard (AES) decision tree 107
170–2 algorithm 74
Data Encryption Standard (DES) 170 classification algorithm 84
private key cryptography 169 classifier 390
public key cryptography 170 de-duplication scheme 127–8
cyber crime 105–9 deep learning 284–6, 330–1
algorithm 310
in healthcare 267
‘data accountability and auditing’ image processing 287–8
model 179–80 for lung disease detection 298–9
data backup 117 defuzzification 225–6
Data Encryption Standard (DES) 170 delta-based de-compression
data mining 62 scheme 128
application 68–9 design verification technologies 40, 46
in banking and finance 68 Destination Sequenced Distance Vector
correlation attribute evaluator (CAE) (DSDV) 238, 242–3
3, 61, 70, 76, 80, 81, 85, 86 Detection of Interest Flooding
in customer segmentation 68 (DIAIF) 357
in distributed architecture 69 differential backup strategy 120, 121
in education 69 digital forensics
educational data mining (EDM) Big Data Analytics in 108–6
(see Educational data mining digital image processing 281–3
(EDM)) digital recovery 119
in fraud detection 68 Direct Memory Access (DMA)
gain ratio attribute evaluator (GRAE) controller 123–4
3, 61, 63, 70, 77, 80, 83–6 Direct Programming Interface (DPI) 37,
in healthcare 68 49–51
information gain attribute evaluator discrete cosine transform (DCT) 194
(IGAE) 3, 61, 63, 70, 80, 82, distributed data processing 17–18
85, 86 Distributed denial-of-service
in multimedia 68–9 (DDoS) 357
operational optimization 68 Dropbox 125
sales and marketing 68 DS18B20 temperature sensor 217
data preprocessing phase 77–8 duplex communication system 309–10
data recovery 118 Dynamic Source Routing (DSR)
Decision Support in Heart Disorder 238, 240
Identification Network dynamic verification 36
(DSHDPS) 341
decision table (DT) 3, 61, 63, 66,
70, 71, 73, 78, 79, 80, 107, E-201C pH sensor 216
330, 341 ECG sensor (AD-8232) 313
404  Intelligent network design driven by Big Data analytics

edge computing 16, 17, 263 Gain ratio attribute evaluator (GRAE)
advantages 261–2 3, 61, 63, 70, 77, 80, 83–6
challenges 272–3 global variables 24
in Covid-19 pandemic era 271–2 Greedy-based Service Orientation
disadvantages 262 Algorithm 139
in healthcare 264–5, 268–9 GSM/General Packet Radio Service
modelling and simulation tools (GPRS) module 309
269–71 GSM/GPRS module 317, 319
Educational data mining (EDM) gyroscope sensor 314
62–5, 70
applications 72
association rule mining algorithm 67 Hadoop Distributed File system
classification algorithms 66 (HDFS) 18, 109
clustering algorithms 66–7 hard drive recovery 118
process 65–6 Hardware Verification Languages
Efficient routing grounded on taxonomy (HVLs) 42
(ERGOT) 122 healthcare
Electronic Design Automations cloud computing in 258–61
(EDAs) 36 deep learning algorithms in 267
Enriched Genetic Algorithm edge computing in 264–5, 268–9
(EGA) 126 IoT in 263–4, 268–9
Equifax (EFX) 28 machine learning in 265–9
equivalent image processing 280 heartbeat sensor 317, 318, 346
ESP32 microcontroller 217, 219 heart diseases 329–30
ETSI-G5 technology 385 deep learning 330–1
Exponential weighted moving average IoT-based heart disease prediction
(EWMA) 361 system
advantages 350
Arduino UNO 345
facial recognition 22 heartbeat sensor 346
Fast Forward Motion Picture Experts limitations 350
Group (FFMPEG) 125 liquid crystal display (LCD)
Field Programmable Gate Array display 347, 348
(FPGA) 35, 36, 39–45 pressure sensor 347, 348
File System Tracker 125 related work 332–4
flutter-based android app 387, 389 results 350–1
Formal verification 36 temperature sensor 347
FPGA: See Field Programmable Gate risk factors 331
Array (FPGA) symptoms 330, 331
Fraud detection 64 types 330
full backup strategy 120 heart rate monitor sensor
Fuzzy inference system (FIS) 213 (MAX 30105) 313
Fuzzy logic controller (FLC) 6, High-security distribution and rake
213, 214, 220, 226, 227, technology (HS-DRT) 122
228, 230 humidity sensor (DHT11) 313
 Index  405

hybrid cloud 166, 260 machine learning (ML) 21–2, 26

Hypertext Transfer Protocol 19 programming 23–4
security 28–31
technology to design a model 24–5
IBN 26 electronic toolkits 25
I2C protocol 41 programming resources 26
iFogSim 270, 271 Intelligent Traffic and Transportation
IGAE: See Information gain attribute Management System (TMS)
evaluator (IGAE) challenges 372–8
image forensics 107 definition 372
image processing 279–80 issues 372–8
color models, applications of 281 machine learning
components 283–4 flutter-based application 387–9
deep learning techniques 287–8 models 388–91
digital image processing 281 for road condition detection
applications 281 378–85
fundamental steps 282–3 that AI approaches 391–4
equivalent 280 intent-based network 22–3, 26
medical imaging 284 Interest flooding attacks (IFA) 9, 357,
virtual image 281 358, 360, 361, 363, 364, 367–9
incremental backup strategy 120, 121 generation, in NDN android 359–60
Information gain attribute evaluator interstitial lung disease detection
(IGAE) 3, 61, 63, 70, 76–7, convolutional neural network (CNN)
80–3, 85, 86 293–5
Information-centric networking
(ICN) 357
benefits 358 JRip
communication model 359 algorithm 3, 61, 75, 78, 79
infrastructure as a service model (IaaS) classification algorithm 86
model 164–5
intelligent devices 14 K-means algorithm 71
connection with networks 14–15 K-means clustering 106, 298, 383, 384
data generation, huge volume of K-nearest neighbor (KNN) 67, 70, 96,
15–18 341, 342, 389–90
intelligent networks algorithm 342
artificial intelligence (AI) 21, 26 classifier 381
Big Data knowledge discovery in databases
cloud and computing 16–17 (KDD) 66
distributed processing 17–18
edge computing 16, 17
business challenges and opportunities language-independent programs 23
26–8 Least Recently Used (LRU) algorithm
evolution 1–15 128
IBN 26 Linear Feedback Shift register (LFSR)
intent-based networking 22–3 47
406  Intelligent network design driven by Big Data analytics

linear regression algorithm 75 android

Linux box 122 application on 364–8
Liquid crystal display (LCD) display attack detection, using AI 368–9
318, 319, 347, 348 communication establishment
Load-balancing framework 141–7 through 363–4
local variables 24 Neptune filtration 127–8
logistic regression (LR) 3, 61, 106, 349, Nericell 378, 383
389, 392 Network AniMator (NAM) 249, 250
lung disease detection Network on Chip (NoC) 41
deep learning for 298–9 neural networks 67, 107
Nio technology 139
NI-my RIO processor 312
machine learning Node MCU ESP 8266 module 312
flutter-based application 387–9
in healthcare 265–9
models 388–91 OFDM system 243–4
in networks 21–2, 26 on-demand self-service 162
for road condition detection 378–85 OpenCourseWare (OCW) 26
Maker Media 25 OpenFlow 141
MapReduce 18, 102–3, 103–5, 106–7 OpenFlow Least-Connections
measured service 162 algorithms 138
medical imaging 284 OpenFlow Round-Robin (RR)
Meet-In-The-Middle (MITM) algorithm 138
technique 172 optical recovery 119
metric-driven verification 40
mined data 96
Modified Bully algorithm 145, 151 packet count 360
Multi Feature(MF)-based Adaptive packet loss 151–2
Neuro-Fuzzy Inference System Packet transmission ratio (PTR)
(ANFIS) 361 151–2
Multilayer perceptron (MLP) 3, 61, Parity cloud services 122
78–80, 83, 296 patient care 269
algorithm 74, 82 Pending interest table (PIT)
classification algorithm 85 utilization 361
Multiple-input multiple-output (MIMO) pH sensor 218
236–8, 245 platform as a service model (PaaS)
model 163–4
Pothole Patrol System 378, 383, 384
Naïve Bayes (NB) POX tool 150
algorithm 3, 61, 73, 78, 79, 80, 84, pressure sensor 347, 348
341, 342, 349, 381 privacy cheating attack model 177
classification 71 private cloud 165–6, 260
classification algorithm 84 private key cryptography 169
NDN Forwarding Daemon (NFD) Pseudo random number generator
362–3 (PRNG) 47, 49–51
 Index  407

public cloud 259 SDN-based load-balancing (SBLB)

security issues in 166 strategy 138
public key cryptography 170 ‘SecCloud’ protocol 176–9
pulse rate body temperature sensor Secure computation confidence
(MAX 30102) 314 (SCC) 177
Python 24, 26, 50, 51 Secure Data Sharing on Cloud
(SeDaSC) model 174–6
Secure storage confidence (SSC) 177
radio frequency identification (RFID) security 28
44, 314 challenges 28–9
rail transport 375 defending corporate world 30–1
random forest 3, 10, 61, 71, 73, 78, 79, end-user devices 29–30
80, 84, 107, 341, 371, 392 personal data and devices 31
algorithm 73–4 safety, needs for 30
classification algorithm 84 Wi-Fi 29
classifier 390–1 Seed Block Algorithm (SBA) 120–1
ranker search method 79–84 sensor-enabled IoT devices 19
Ransomware Attack Risk Indicator sentiment analysis 70
(RARI) hardware module 123–4 Serial plotter software 309
rapid elasticity 161–2 service-oriented business model 117
Rapid Prototyping Method 25 Simulated Annealing (SA)-based
Raspberry Pi 25, 312 Service Orientation
Raspberry Pi-based health monitoring algorithm 139
system 311 smart health-care systems 303–5
RD Iterative Adaptive Inverse Filtering development boards 312
(RD-IAIF) 360–2 GSM/GPRS module 314–15
real-time edge analytics hardware components 316–20
for Big Data 109–11 implementation 320–3
Red, Green, Blue (RGB) model 281 literature survey 305–8
Redundant Array of Inexpensive radio frequency identification
Devices (RAID) recovery 119 (RFID) 314
Register Transfer Level (RTL) 35, 36, 39 related works 309–11
regression 61, 63, 66, 96 results 320–3
removable recovery 119 sensors 312–14
resource pooling 162 workflow 315–16
respiration sensor 314 smartphones 117
RESTful application programming sniffer attacks 168
interfaces (APIs) 19 SoC verification 44, 45
Root replies packets (RREPs) 238 soft-computing techniques
RREQ 238, 239 speaker identification 196
software as a service (SaaS) model
162, 163
satisfaction ratio 361 software defined accelerator
SDN: See Software-defined networking (SODA) 43
(SDN) Software-defined networking (SDN)
408  Intelligent network design driven by Big Data analytics

architecture overview 140–1 Takagi–Sugeno system 361

controller architectures tape recovery 119
centralized architecture 144 taxi-based mobile sensing system 384
distributed architecture 142–4 Tcl 51
hybrid architecture 144 temperature sensor 219, 313, 347,
controller head selection strategy 316–17
144–6 ThingSpeak 227–9
load-balancing framework 141–2 threat management techniques
Modified Bully algorithm 151 cryptography
overview 135, 137, 138 Advance Encryption Standard
POX tool 150 (AES) 170–2
super controller Data Encryption Standard
implementation framework 148 (DES) 170
multiple controllers 149 private key cryptography 169
network configuration 148 public key cryptography 170
Solid-State Drives (SSD) 123 RSA algorithm 172–4
speaker recognition system Trinity College Dublin Speaker
feature extraction 192–5 Aging (TCDSA) database 190
feature mapping 195–6 turbidity sensor 219
human speech production
system 197
overview 185–7 Universal Verification Methodology
speaker database collection 187–92 (UVM) 3, 35–9, 41–5, 52,
speaker localization 203 53, 55
speech signals UVM: See Universal Verification
autocorrelation 200 Methodology (UVM)
perceptual features 202
physical features 200–1
signal features 320–203 variables 24
sinusoids, properties of 198–9 verification management, VLSI
speech perception 198 case study 46–7, 51–3
speech production system 197–8 PRNG implementation 49–51
windowing signals 199 pseudo random number generator
zero-crossing rate (ZCR) 199 47–9
speech production system 197–8 challenges 54–5
speech recognition 21 environment root block
spoofing attacks 168, 190 structure 38
storage cheating attacks 177 vibration sensor 314, 317, 318
super controllers 145, 146, 148, 149 Virtual Machine Monitor (VMM)
System Initialization Operator 165, 167
(SIO) 178 virtual machines (VMs) 18
system Verilog 3, 35, 41, 43, 44, 50, 52 VLSI computing system
SystemVerilog Direct Programming intelligent verification management
Interface in this work approach (see verification
(SV-DPI) 42 management, VLSI)
 Index  409

Water quality-monitoring and decision- Wireless sensor networks (WSN) 235,

making system (WQMDMS) 245–6
experimental procedures 217–20 application 240, 246–7
overview 213–14 future scope 252–3
results 227–30 limitations 247
working 220–7 simulation parameters
Web scraping software 19 248–50
Wi-Fi module (ESP8266) 319
Wi-Fi security 29 xorshift algorithm 48
WiMAX 243 xorshift network 49
WiMAX-based WSN system 249–50
windowing signals 199 zero-crossing rate (ZCR) 199
Intelligent Network Design Driven by Big Data
Analytics, IoT, AI and Cloud Computing

Intelligent Network Design

As enterprise access networks evolve with a larger number of mobile users, a wide range of

Analytics, IoT, AI and Cloud Computing

Intelligent Network Design Driven by Big Data
devices and new cloud-based applications, managing user performance on an end-to-end
basis has become rather challenging. Recent advances in big data network analytics combined
with AI and cloud computing are being leveraged to tackle this growing problem. AI is
becoming further integrated with software that manage networks, storage, and can compute. Driven by Big Data Analytics,
IoT, AI and Cloud Computing
This edited book focuses on how new network analytics, IoTs and Cloud Computing platforms
are being used to ingest, analyse and correlate a myriad of big data across the entire network
stack in order to increase quality of service and quality of experience (QoS/QoE) and to
improve network performance. From big data and AI analytical techniques for handling the
huge amount of data generated by IoT devices, the authors cover cloud storage optimization,
the design of next generation access protocols and internet architecture, fault tolerance and
reliability in intelligent networks, and discuss a range of emerging applications.

This book will be useful to researchers, scientists, engineers, professionals, advanced students Edited by
and faculty members in ICTs, data science, networking, AI, machine learning and sensing. It
will also be of interest to professionals in data science, AI, cloud and IoT start-up companies, Sunil Kumar, Glenford Mapp and Korhan Cengiz
as well as developers and designers.

About the Editors

Sunil Kumar is an associate professor of computer science and engineering at Amity
University, Noida campus, India.

Glenford Mapp is an associate professor at Middlesex University, London, UK.

Korhan Cengiz is an assistant professor of electrical and electronics engineering at Trakya

University, Turkey.

Edited by Kumar, Mapp and Cengiz

The Institution of Engineering and Technology

theiet.org
978-1-83953-533-8