HANDWRITTEN DIGIT RECONIZATION USING

MACHINE LEARNING
A PROJECT REPORT

Submitted by

NAME UID
HARSH HARSHIT 21BCS9622

BRIJESH KISHORE PURPOIT 21BCS8816

RAHUL YADAV 21BCS9479

MIRNAL BHATT 21BCS9076

in partial fulfillment for the award of the degree of

BACHELOR OF ENGINEERING
IN

COMPUTER SCIENCE (HONS) ARTIFICAL INTELLIGENCE AND

MACHINE LEARNING

Chandigarh University
OCT 2023

1
 BONAFIDE CERTIFICATE
Certified that this project report “Handwritten digit recognition using machine
learning” is the bonafide work of “Harsh Harshit, Rahul Yadav, Brijesh Kishore
Purohit, Mrinal Bhatt” who carried out the project work under my supervision Prabjot
Singh Bali.

SIGNATURE SIGNATURE

-------------------
------------------

<<Department> <<Academic Designation>>

>

INTERNAL EXAMINER EXTERNAL EXAMINER

2
 Acknowledgment

We extend our sincere gratitude to all those who have contributed to the successful completion of our project on
"Handwritten Digit Recognition Using Machine Learning." This endeavor would not have been possible without
the support, guidance, and encouragement of numerous individuals and organizations. First and foremost, we
express our deepest appreciation to our supervisor [Supervisor's Name], whose unwavering guidance and valuable
insights steered us throughout this project. Their expertise in the field of machine learning, particularly in
Convolutional Neural Networks (CNN), OpenCV, Support Vector Machines (SVM), and image classification, has
been instrumental in shaping our understanding and approach. We are immensely thankful to the faculty members
of Prabjot Singh Bali. for their continuous encouragement and support. Their feedback and suggestions have been
invaluable in refining our methodologies and enhancing the quality of our work. We would like to acknowledge the
contributions of our classmates and colleagues who provided assistance and constructive feedback during various
stages of the project. Their diverse perspectives and collaborative spirit have enriched our learning experienc e. Our
heartfelt appreciation goes to the developers of the open-source libraries and frameworks that formed the backbone
of our project. We are indebted to the creators and contributors of CNN models, OpenCV functionalities, SVM
implementations, and image classification algorithms for providing accessible tools and resources that enabled us
to explore innovative solutions. Furthermore, we extend our gratitude to the participants who generously shared
their handwritten digit datasets, allowing us to conduct experiments and evaluate our models effectively. Last but
not least, we express our profound gratitude to our families and friends for their unwavering support and
understanding throughout this journey. Their encouragement and belief in our abilities have been a constant source
of motivation.

3
 Abstract

This project focuses on Handwritten Digit Recognition using machine learning techniques, specifically
Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs). The primary dataset employed
is the MNIST Dataset, a benchmark in handwritten digit recognition. Additionally, the project incorporates a mix
of supplementary datasets to enhance model robustness. Leveraging the power of CNNs, the system extracts
hierarchical features from digit images, allowing for improved classification accuracy. SVMs complement the
approach by providing a robust classification algorithm. The implementation utilizes OpenCV for image
preprocessing, ensuring optimal input for the machine learning models. Through a combination of transfer
learning and feature engineering, the system achieves superior performance in accurately identifying handwritten
digits. The fusion of diverse datasets enables the model to generalize well across various writin g styles and
conditions. The project aims to showcase the effectiveness of a hybrid approach, combining the strengths of
CNNs and SVMs, for handwritten digit recognition. The experimental results demonstrate the potential of this
methodology for real-world applications in digitized document processing, automated form recognition, and
other domains where handwritten digit recognition is crucial.

4
TABLE OF CONTENTS
1. Introduction ....................................................................................................... 7
1.1 Identification of client & need.......................................................................... 8
1.2 Relevant contemporary issues .......................................................................... 9
1.3 Problem Identification..................................................................................... 10
1.4 Task Identification........................................................................................... 11
1.5 Timeline ........................................................................................................... 13
2. Literature survey.............................................................................................. 14
2.1 Timeline of the reported problem as investigated the world… ...................... 16
2.2 Bibliometric analysis ........................................................................................ 17
2.3 Proposed solutions by different researchers ................................................... 18
2.4 Summary linking literature review with the project ...................................... 22
2.5 Objective ........................................................................................................... 26
2.6 Problem Definition ............................................................................................ 28
3. Design Process .................................................................................................. 30
3.1 Concept Generation ......................................................................................... 31
3.2 Evaluation & Selection of Specifications ........................................................ 33
3.3 Design Constraints ........................................................................................... 34
3.4 Analysis and Feature finalization .................................................................... 35
3.5 Design Flow....................................................................................................... 37
3.6 Best Design selection......................................................................................... 38
3.7 Implementation plan ......................................................................................... 39
4. Results analysis and validation ........................................................................ 40

5
4.1 Implementation of design................................................................................. 47
4.2 , Design Drawings ............................................................................................. 49
4.3 Report preparation........................................................................................... 51
4.4 Train, Test Validation of Data .......................................................................... 52
4.5 Accuracy of Data.............................................................................................. ,54
5. Conclusion and Future Work .......................................................................... 56
5.1 Future Scope ..................................................................................................... 57
5.2 Conclusion ........................................................................................................ 58
5.3 Summary........................................................................................................... 59
5.4 Reference .......................................................................................................... 60

6
1. Introduction:
Handwritten digit recognition stands as a cornerstone problem in the field of pattern recognition and
machine learning. It serves as a fundamental task with broad applications ranging from postal
automation to digital document processing. The ability to accurately identify handwritten digits holds
immense significance in fields such as finance, healthcare, and education, where numerical data often
requires efficient and reliable processing. In recent years, with the advancements in machine learning
techniques, particularly in Convolutional Neural Networks (CNNs), the accuracy and efficiency of
handwritten digit recognition systems have witnessed remarkable improvements. This progress has
been fueled by the availability of large-scale digit datasets, such as MNIST (Modified National
Institute of Standards and Technology), and the development of sophisticated algorithms capable of
extracting discriminative features from raw pixel data. The aim of this project is to explore and
implement a robust handwritten digit recognition system using machine learning techniques,
specifically employing CNNs, OpenCV, Support Vector Machines (SVMs), and image classification
methodologies. By leveraging these powerful tools and methodologies, we seek to develop a system
that can accurately classify handwritten digits with high precision and efficiency. Convolutional
Neural Networks (CNNs) have emerged as the state-of-the-art approach for image recognition tasks
due to their ability to automatically learn hierarchical features from raw input data. CNNs excel in
capturing spatial dependencies and local patterns within images, making them well -suited for tasks
like handwritten digit recognition. OpenCV (Open Source Computer Vision Library) serves as a
versatile tool for image processing and computer vision tasks. In this project, we utilize OpenCV to
preprocess raw images, extract relevant features, and enhance the quality of input data before feeding
it into the machine learning models. Support Vector Machines (SVMs) offer a robust framework for
classification tasks by identifying the optimal hyperplane that separates different classes in the feature
space. SVMs have been widely used in conjunction with CNNs for classification tasks, providing a
complementary approach to improve overall accuracy and generalization. Image classification, a
fundamental task in computer vision, involves assigning predefined labels or categories to images
based on their visual content. In this project, we employ image classification techniques to assign
labels to handwritten digits, thereby enabling the recognition system to accurately identify and
classify digits from input images. Through this project, we aim to not only develop a functional
handwritten digit recognition system but also contribute to the ongoing research and development in
7
the field of machine learning and computer vision. By harnessing the power of CNNs, OpenCV,
SVMs, and image classification techniques, we strive to push the boundaries of accuracy and
efficiency in handwritten digit recognition, ultimately paving the way for innovative applications in
various domains.

1.1 Identification of client & need

The client for this project encompasses various stakeholders across industries where efficient digit
recognition plays a pivotal role. These stakeholders include financial institutions, postal services,
educational institutions, healthcare providers, and any entity dealing with handwritten numerical data.
The primary need addressed by this project is the accurate and efficient recognition of handwritten
digits. In many sectors, handwritten digits are prevalent, whether in forms, documents, checks, or
postal addresses. The manual processing of such data is time-consuming, error-prone, and resource-
intensive. Therefore, there is a pressing need for automated systems that can swiftly and accurately
recognize handwritten digits, streamlining processes and enhancing overall efficiency. Moreover, the
demand for digit recognition systems stems from the increasing digitization of various industries. As
organizations transition towards digital workflows, the ability to extract and interpret handwritten
numerical data becomes indispensable. From automating financial transactions to sorting mail, digit
recognition systems serve as a critical component in modernizing operations and improving
productivity. Additionally, there is a growing emphasis on machine learning and artificial intelligence
technologies in addressing real-world challenges. Handwritten digit recognition serves as an ideal
domain for applying these advanced techniques, providing a tangible and practical application of
cutting-edge research in pattern recognition, image processing, and classification algorithms. By
identifying the client as industries reliant on handwritten numerical data and recognizing their need
for accurate and efficient digit recognition, this project aims to develop a solution that not only meets
the immediate requirements but also aligns with the broader trends towards automation and
digitization. Through the implementation of machine learning techniques, such as CNNs, OpenCV,
SVMs, and image classification methodologies, the project endeavors to deliver a robust and
adaptable digit recognition system capable of addressing the diverse needs of clients across various
sectors.

8
1.2 Relevant contemporary issues

In the realm of handwritten digit recognition and machine learning, several contemporary issues shape
the landscape and influence the development of innovative solutions. Understanding these issues is
crucial for addressing challenges effectively and devising strategies that align with current trends and
advancements. Some of the key contemporary issues include:

1. Data Privacy and Security: With the increasing reliance on digitization and machine learning
algorithms, ensuring the privacy and security of sensitive data has become paramount. Handwritten
digit recognition systems often deal with personal information, financial data, and confidential
documents. Addressing concerns regarding data privacy, protecting against unauthorized access, and
implementing robust security measures are essential considerations in the development and
deployment of such systems.

2. Bias and Fairness: Machine learning algorithms are susceptible to bias, which can lead to unfair
outcomes, particularly in areas such as automated decision-making and classification. Addressing
bias in handwritten digit recognition systems involves careful examination of training data,
algorithmic fairness, and mitigation strategies to ensure equitable treatment across diverse
demographics.

3. Interpretability and Explainability: As machine learning models become more complex, there is a
growing need for interpretability and explainability. Stakeholders require insights into how models
make predictions, particularly in high-stakes applications such as healthcare and finance. Enhancing
the interpretability of handwritten digit recognition systems promotes trust, accountability, and

9
understanding among users and decision-makers.

4. Ethical Considerations: Ethical considerations surrounding the use of machine learning algorithms
extend beyond bias and fairness to encompass broader societal impacts. Handwritten digit recognition
systems should adhere to ethical principles such as transparency, accountability, and fairness.
Addressing ethical dilemmas, such as the potential misuse of digit recognition technology or its
unintended consequences, requires proactive measures and ethical frameworks to guide development
and deployment.

5. Continual Learning and Adaptation: Handwritten digit recognition systems must be capable of
continual learning and adaptation to evolving data distributions, user preferences, and environmental
factors. The ability to adapt to new handwriting styles, languages, or input modalities ensures the
long-term viability and effectiveness of these systems in real-world scenarios.

6. Robustness to Adversarial Attacks: Machine learning models, including those used in handwritten
digit recognition, are susceptible to adversarial attacks, where small perturbations to input data can
lead to incorrect predictions. Enhancing the robustness of digit recognition systems against such
attacks involves developing resilient models, robust training methodologies, and validation
techniques that account for potential vulnerabilities.

By addressing these contemporary issues in the development and deployment of handwritten digit
recognition systems, stakeholders can foster trust, promote fairness, and ensure the responsible use
of machine learning technology in diverse applications. Moreover, proactive engagement with these
issues enables the advancement of ethical and equitable solutions that align with societal values and
expectations.

1.3 Problem Identification

The problem at hand revolves around the inefficiency and error-proneness inherent in manual
processing of handwritten numerical data. Despite the widespread digitization of various sectors,
handwritten digits persist in forms, documents, and other sources, posing a challenge for automated

10
data processing systems.

One primary issue is the variability and inconsistency in handwritten digits, stemming from
differences in writing styles, stroke thickness, and individual nuances. This variability makes it
challenging to develop accurate and robust recognition algorithms that can generalize well across
diverse handwriting samples.

Moreover, manual digit recognition processes are labor-intensive and time-consuming, leading to
bottlenecks in workflows and hindering overall productivity. Human errors in transcribing
handwritten digits further exacerbate these challenges, potentially leading to inaccuracies, delays, and
financial losses in critical applications such as financial transactions or postal sorting.

Existing digit recognition systems often rely on simplistic approaches or traditional machine learning
algorithms, which may lack the sophistication and adaptability needed to handle real-world variability
effectively. As a result, these systems may struggle with accuracy, particularly when confronted with
handwritten digits exhibiting unconventional styles or variations.

Furthermore, the rapid pace of digitization and technological advancements necessitates the
development of scalable and adaptable digit recognition solutions capable of handling large volumes
of data, diverse input formats, and evolving user requirements.

Therefore, the overarching problem is to design and implement a robust, accurate, and efficient
handwritten digit recognition system that addresses the challenges of variability, scalability, accuracy,
and adaptability. This system should leverage advanced machine learning techniques, such as
Convolutional Neural Networks (CNNs), OpenCV, Support Vector Machines (SVMs), and image
classification methodologies, to achieve high levels of accuracy and efficiency across diverse
handwriting samples and applications.

By identifying and addressing these key challenges, the goal is to develop a solution that streamlines
digit recognition processes, enhances productivity, reduces errors, and facilitates the seamless
integration of handwritten numerical data into digital workflows across various industries.

11
1.4 Task Identification

The task at hand involves a series of interconnected steps aimed at developing and implementing a
robust handwritten digit recognition system. These tasks encompass various stages of data
preprocessing, model development, training, evaluation, and deployment. The key tasks involved in
this project include:

1. Data Collection: Gather a diverse dataset of handwritten digits, encompassing different writing
styles, variations, and backgrounds. Popular datasets such as MNIST may be used, along with
additional custom datasets to enhance model generalization.

2. Data Preprocessing: Clean and preprocess the collected data to standardize formats, remove noise,
and enhance readability. Techniques such as normalization, resizing, and noise reduction may be
applied to ensure consistency and improve model performance.

3. Feature Extraction: Extract relevant features from preprocessed images to facilitate digit
recognition. This may involve techniques such as edge detection, contour extraction, and texture
analysis to capture discriminative information from handwritten digits.

4. Model Selection: Choose appropriate machine learning models for digit recognition, considering
factors such as model complexity, computational efficiency, and performance metrics. Convolutional
Neural Networks (CNNs), Support Vector Machines (SVMs), and ensemble methods may be
explored for their suitability in this task.

5. Model Development: Develop and train the selected models using the preprocessed data. This
involves defining model architectures, specifying hyperparameters, and optimizing performance
metrics such as accuracy, precision, and recall through iterative training and validation.

6. Model Evaluation: Evaluate the trained models using separate test datasets to assess their
performance and generalization capabilities. Metrics such as accuracy, confusion matrices, and
receiver operating characteristic (ROC) curves may be employed to quantify model performance and

12
identify areas for improvement.

7. Hyperparameter Tuning: Fine-tune model hyperparameters to optimize performance and enhance

robustness. Techniques such as grid search, random search, and Bayesian optimization may be
utilized to explore the hyperparameter space and identify optimal configurations.

8. Deployment and Integration: Deploy the trained model into production environments, integrating
it with existing systems or applications where handwritten digit recognition is required. This may
involve developing APIs, web services, or standalone applications to facilitate seamless integration
and real-time inference.

9. Testing and Validation: Conduct rigorous testing and validation of the deployed system to ensure
reliability, scalability, and usability in real-world scenarios. This includes stress testing, performance
monitoring, and user acceptance testing to validate system functionality and performance under
diverse conditions.

10. Documentation and Reporting: Document the entire development process, including data
collection, preprocessing steps, model architectures, training procedures, and evaluation results.
Prepare comprehensive reports and documentation to facilitate knowledge transfer, collaboration, and
future iterations of the digit recognition system.

By systematically addressing these tasks, the project aims to develop a robust and accurate
handwritten digit recognition system that meets the needs of various stakeholders across industries,
streamlining digit recognition processes and enhancing productivity in diverse applications.

13
1.5 Timeline:
Project Planning and Dataset Acquisition (2 weeks):

Define project objectives, scope, and requirements.

Identify and acquire the dataset for handwritten digits.

Establish data preprocessing requirements.

Data Preprocessing (2 weeks):

Clean the dataset by removing noise, outliers, and inconsistencies.

Normalize pixel values and standardize image dimensions.

Augment the dataset if necessary to increase variability.

Model Development and Training (4 weeks):

Research and select appropriate model architecture (e.g., CNN).

Implement the model using a deep learning framework like TensorFlow or PyTorch.

Split the dataset into training, validation, and test sets.

Train the model on the training data, tuning hyperparameters as needed.

Validate the model's performance on the validation set and fine-tune as necessary.

Model Evaluation and Optimization (2 weeks):

Evaluate the trained model's performance on the test set.

14
Analyze metrics such as accuracy, precision, recall, and F1 score.

Identify areas for improvement and optimize the model accordingly.

Implement regularization techniques to prevent overfitting if necessary.

Deployment Preparation (2 weeks):

Prepare the model for deployment, including serialization and containerization.

Develop APIs or interfaces for integrating the model into production systems.

Test the deployment environment and ensure compatibility with infrastructure.

Deployment and Monitoring (2 weeks):

Deploy the model into the production environment.

Monitor model performance in real-world scenarios.

Implement logging and error handling mechanisms for troubleshooting.

Continuously monitor and update the model to maintain accuracy and reliability.

Documentation and Knowledge Transfer (1 week):

Document the project including code, model architecture, and deployment instructions.

Prepare user manuals or guides for stakeholders.

Conduct knowledge transfer sessions with relevant team members or stakeholders.

15
2. Literature survey

A comprehensive literature survey serves as the cornerstone for understanding the current state -of-
the-art, identifying research gaps, and informing the development of novel solutions in handwritten
digit recognition using machine learning techniques. In this section, we review key studies,
methodologies, and advancements in the field:

1. Convolutional Neural Networks (CNNs) for Handwritten Digit Recognition: CNNs have emerged
as the dominant approach for digit recognition tasks due to their ability to automatically learn
hierarchical features from raw input data. LeNet-5, proposed by LeCun et al. (1998), was one of the
pioneering CNN architectures for handwritten digit recognition, achieving impressive performance
on the MNIST dataset. Since then, various CNN architectures, including AlexNet, VGG, ResNet, and
MobileNet, have been adapted and optimized for digit recognition tasks, leading to significant
improvements in accuracy and efficiency.

2. Data Augmentation Techniques: Data augmentation plays a crucial role in mitigating overfitting
and improving the generalization capabilities of digit recognition models. Techniques such as
rotation, translation, scaling, and elastic deformation have been widely used to generate synthetic
training samples, effectively expanding the diversity and variability of the training dataset. Studies
by Simard et al. (2003) and Krizhevsky et al. (2012) have demonstrated the effectiveness of data
augmentation in enhancing model robustness and performance.

3. Transfer Learning and Fine-tuning: Transfer learning, a technique where pre-trained models are
adapted to new tasks, has gained prominence in digit recognition tasks. By leveraging features learned
from large-scale image datasets such as ImageNet, pre-trained CNNs can be fine-tuned on smaller

16
handwritten digit datasets, leading to faster convergence and improved performance. Research by
Yosinski et al. (2014) and Razavian et al. (2014) has highlighted the efficacy of transfer learning in
digit recognition and other computer vision tasks.

4. Ensemble Learning Approaches: Ensemble learning methods, such as bagging, boosting, and
stacking, have been employed to enhance the accuracy and robustness of digit recognition models.
By combining multiple base learners, ensemble methods can effectively capture diverse patterns and
reduce model variance, leading to improved generalization performance. Studies by Dietterich (2000)
and Freund and Schapire (1997) have demonstrated the effectiveness of ensemble learning in
handwritten digit recognition and other classification tasks.

5. SVMs and Kernel Methods: Support Vector Machines (SVMs) and kernel methods have been
widely used in conjunction with CNNs for digit recognition tasks. SVMs offer a robust framework
for classification tasks by identifying optimal decision boundaries in high-dimensional feature spaces.
Studies by Cortes and Vapnik (1995) and Burges (1998) have explored the effectiveness of SVMs in
digit recognition, particularly when combined with handcrafted features or deep learning embeddings.

6. Real-world Applications and Challenges: Beyond academic benchmarks such as MNIST,

handwritten digit recognition finds applications in various real-world scenarios, including postal
automation, document processing, bank check recognition, and digitized archival systems. However,
challenges such as variability in handwriting styles, noise in scanned documents, and scalability to
large datasets pose significant hurdles in deploying effective digit recognition systems in practical
settings.

7. Ethical and Societal Implications: The deployment of digit recognition systems raises ethical and
societal considerations regarding privacy, bias, fairness, and accountability. Ensuring transparency,
fairness, and responsible use of digit recognition technology are essential to mitigate potential harms

17
and promote trust among users and stakeholders.

By synthesizing insights from these studies and methodologies, this literature survey provides a
foundational understanding of the principles, techniques, and challenges in handwritten digit
recognition using machine learning. Building upon this knowledge, the subsequent sections of this
project will delve into the design, implementation, and evaluation of a novel digit recognition system
that addresses the identified research gaps and advances the state-of-the-art in the field.

2.1 Bibliometric analysis:

Bibliometric analysis involves quantitatively analyzing the scholarly literature to identify trends,
patterns, and key contributors within a particular field of study. In the context of handwritten digit
recognition, a bibliometric analysis can provide insights into the volume of research, citation patterns,
influential authors, top journals, and emerging trends. Below, I'll outline the steps involved in
conducting a bibliometric analysis:

1. Literature Retrieval: Collect relevant scholarly articles, conference papers, and other publications
related to handwritten digit recognition. This can be done using academic databases such as PubMed,
IEEE Xplore, Google Scholar, Scopus, or Web of Science. Include keywords such as "handwritten
digit recognition," "digit recognition," "convolutional neural networks," "machine learning," etc.

2. Data Preparation: Compile the retrieved literature into a bibliographic database, ensuring
consistency in formatting and metadata. Organize the data according to publication year, author(s),
journal/conference, citation count, keywords, and other relevant attributes.

18
3. Data Analysis: Utilize bibliometric software tools or programming languages (e.g., Python with
bibliometrics packages) to analyze the dataset. Common bibliometric indicators include:

- Publication Trends: Plot the number of publications over time to visualize the growth of research
in handwritten digit recognition.

- Authorship Patterns: Identify prolific authors and collaboration networks through co-authorship
analysis.

- Journal/Conference Analysis: Determine the most influential journals or conferences publishing

research on handwritten digit recognition.

- Citation Analysis: Examine citation patterns to identify highly cited papers and influential research
contributions.

- Keyword Analysis: Analyze keyword frequencies to identify prevalent topics and emerging trends
within the field.

- Institutional Analysis: Identify leading institutions or countries contributing to research in

handwritten digit recognition.

4. Visualization: Present the findings of the bibliometric analysis using visualizations such as bar
charts, line graphs, network diagrams, and word clouds. Visual representations can enhance the
interpretation and communication of bibliometric insights.

19
5. Interpretation: Interpret the results of the bibliometric analysis to draw conclusions about the state-
of-the-art, research trends, and areas of focus within the field of handwritten digit recognition. Discuss
notable findings, influential authors, emerging topics, and potential future directions for research.

6. Limitations and Considerations: Acknowledge any limitations of the bibliometric analysis, such as
biases in database coverage, variations in citation practices, and the exclusion of non-indexed sources.
Consider the broader context of the findings and their implications for advancing research in
handwritten digit recognition.

By conducting a bibliometric analysis, researchers can gain valuable insights into the landscape of
scholarly literature, identify influential works and authors, and inform future research directions in
the field of handwritten digit recognition.

2.2 Proposed solutions by different researchers:

Rohit Chandra Joshi, Vivek Raj Patel, Anjali Goyal Evaluation of Supervised Machine Learning
Models for Handwritten Digit Recognition 2022 4th International Conference on Advances in
Computing, Communication Control and Networking (ICAC3N). They compare different neural
network approaches based on factors such as accuracy, performance, and execution time, using both
random and standard datasets. [1]

Ashraf Y. Maghari, Mahmoud M. Abu Ghosh A Comparative Study on Handwriting Digit

Recognition Using Neural Networks 2017 International Conference on Promising Electronic
Technologies. Maghari and Abu Ghosh conduct a comparative study on handwriting digit recognition

20
using neural networks. They explore the performance of various neural network models in
recognizing handwritten digits.[2]

Saqib Ali, Zareen Sakhawat, Tariq Mahmood, Muhammad Saqlain Aslam, Zeeshan Shaukat,
Sana Sahiba A robust CNN model for handwritten digits recognition and classification 2020 IEEE
International Conference on Advances in Electrical Engineering and Computer Applications
(AEECA). Ali et al. propose a robust convolutional neural network (CNN) model for handwritten
digit recognition and classification. They focus on improving accuracy in recognizing handwritten
digits, particularly using the MNIST dataset.[3]

Huseyin Kusetogullari, Amir Yavariabdi, Johan Hall, Niklas LavessonA Deep Handwritten Digit
Detection and Recognition Method Using a New Historical Handwritten Digit Dataset Big Data
Research (2021). Kusetogullari, Yavariabdi, Hall, and Lavesson present a deep learning approach for
handwritten digit detection and recognition. They introduce a new historical handwritten digit dataset
and propose a method to improve accuracy in detecting and recognizing handwritten digits.[4]

Jinze Li, Gongbo Sun, Leiye Yi, Qian Cao, Fusen Liang, Yu Sun Handwritten Digit Recognition
System Based on Convolutional Neural Network 2020 IEEE International Conference on
Advances in Electrical Engineering and Computer Applications (AEECA). Li et al. develop a
handwritten digit recognition system based on convolutional neural network (CNN). Their system
aims to enhance efficiency and accuracy in recognizing handwritten digits, particularly focusing on
applications in banking operations.[5]

Authors: Shubham Mendapara, Krish Pabani, Yash Paneliya Handwritten Digit Recognition
System International Journal of Scientific Research in Computer Science, Engineering and
Information Technology. Mendapara, Pabani, and Paneliya propose a handwritten digit recognition

21
system. Their system utilizes machine learning techniques to efficiently process large handwritten
documents and improve recognition accuracy.[6]

Authors: Elizabeth Rani G, Sakthimohan M, Abhigna Reddy G, Selvalakshmi D, Thomalika

Keerthi, Raja Sekar R MNIST Handwritten Digit Recognition using Machine Learning 2022 2nd
International Conference on Advance Computing and Innovative Technologies in Engineering
(ICACITE). Elizabeth Rani et al. explore MNIST handwritten digit recognition using machine
learning algorithms. They evaluate various machine learning classifiers to enhance the efficiency and
accuracy of handwritten digit recognition systems.[7]

Mariam Omar Alshrief, Wafa El-Tarhouni, Kenz Amhmed Bozed Ensemble machine learning
model for classification of handwritten digit recognition 2022 IEEE 2nd International Maghreb
Meeting of the Conference on Sciences and Techniques of Automatic Control and Computer
Engineering (MI-STA) Alshrief, El-Tarhouni, and Bozed introduce an ensemble machine learning
model for the classification of handwritten digit recognition. Their approach aims to improve
classification efficacy by integrating multiple classifiers.[8]

Ayush Kumar Agrawal, A.K. Shrivas, Vineet Kumar Awasthi A Robust Model for Handwritten
Digit Recognition using Machine and Deep Learning Technique 2021 2nd International
Conference for Emerging Technology (INCET) Agrawal, Shrivas, and Awasthi propose a robust
model for handwritten digit recognition using machine and deep learning techniques. Their model
aims to achieve high accuracy in recognizing handwritten digits through machine learning
techniques.[9]

Paras Nath Singh, Kiran Babu T S Recognition of Scanned Handwritten digits using Deep
Learning 2023 IEEE 3rd Mysore Sub Section International Conference (MysuruCon). Singh and

22
Kiran present a deep learning approach for recognizing scanned handwritten digits. Their method
achieves high accuracy in recognizing handwritten digits from scanned images using deep learning
techniques.[10]

Authors: Md. Ferdous Wahid, Md. Fahim Shahriar, Md. Shohanur Islam Sobuj A Classical
Approach to Handcrafted Feature Extraction Techniques for Bangla Handwritten Digit
Recognition 2021 International Conference on Electronics, Communications and Information
Technology (ICECIT) Wahid, Shahriar, and Sobuj propose a classical approach to handcrafted
feature extraction techniques for Bangla handwritten digit recognition. Their method aims to improve
recognition performance using feature extraction techniques.[11]

Xiao Xiao, Bolin Liao, Qiuqing Long, Yongjun He, Jun Li, Luyang Han Improved Extreme
Learning Machine Based on Deep Learning and Its Application in Handwritten Digits Recognition
2021 11th International Conference on Intelligent Control and Information Processing (ICICIP)
Xiao et al. introduce an improved extreme learning machine based on deep learning for handwritten
digits recognition. Their approach aims to enhance classification accuracy using deep learning
techniques.[12]

R Anoop, Niharika Kiran, Bhargav P Raj, Ashwini Kodipalli, Trupthi Rao, B R Rohini MNSIT
Handwritten Digit Recognition Using Machine Learning Classification Algorithms 2023 World
Conference on Communication & Computing (WCONF) Anoop et al. explore MNIST handwritten
digit recognition using machine learning classification algorithms. They assess classification
accuracy with varied configurations and highlight the challenges in accurate digit recognition.[13]

Huihuang Zhao, Han Liu Algebraic Fusion of Multiple Classifiers for Handwritten Digits
Recognition 2018 International Conference on Wavelet Analysis and Pattern Recognition

23
(ICWAPR) Zhao and Liu propose an algebraic fusion of multiple classifiers for handwritten digits
recognition. Their approach aims to enhance recognition accuracy by integrating multiple
classifiers.[14]

Aminul Islam, Fuad Rahman, AKM Shahariar Azad Rabby Sankhya: An Unbiased Benchmark
for Bangla Handwritten Digits Recognition 2019 IEEE International Conference on Big Data (Big
Data) Islam, Rahman, and Rabby introduce Sankhya, an unbiased benchmark for Bangla handwritten
digits recognition. Their benchmarking process aims to improve the evaluation of recognition
performance across publicly available datasets.[15]

Kairui Jin Handwritten digit recognition based on classical machine learning methods 2022 3rd
International Conference on Electronic Communication and Artificial Intelligence (IWECAI) Jin
investigates handwritten digit recognition based on classical machine learning methods. Their study
challenges the dominance of CNN-based approaches, revealing superior image classification
capabilities of classical machine learning methods.[16]

Soha Boroojerdi, George Rudolph Handwritten Multi-Digit Recognition With Machine Learning
2022 Intermountain Engineering, Technology and Computing (IETC) Boroojerdi and Rudolph
investigate handwritten multi-digit recognition with machine learning. Their study reveals Random
Forest as the most accurate method for recognizing multi-digit numbers.[17]

Kumar Priyadarshini, Satnam Singh, Anil Kumar Dixit Recognition of Handwritten Digit Using
Different Machine Learning Algorithms 2022 International Conference on Cyber Resilience
(ICCR). Priyadarshini, Singh, and Dixit address the recognition of handwritten digits using different
machine learning algorithms. Their study aims to enhance efficiency and accuracy in digit
recognition.[18]

24
 Sachin Chauhan, Samar Mahmood, T. Poongodi Handwritten Digit Recognition using Deep
Neural Networks 2023 4th International Conference on Intelligent Engineering and Management
(ICIEM). Chauhan, Mahmood, and Poongodi explore handwritten digit recognition using deep neural
networks. Their study aims to identify the most effective model through accuracy and execution time
evaluations.[19]

Smitha N, Rahul Kumar Singh, Subodh Kumar Yadav, Sandeep Sah, Keshava Kumar N, Hemanth
Kumar H An Empirical Comparison of Handwritten Character Recognition Using Machine
Learning 2021 Second International Conference on Electronics and Sustainable Communication
Systems (ICESC). Smitha et al. conduct an empirical comparison of handwritten character
recognition using machine learning. Their study aims to analyze existing research efforts and explore
potential future directions for enhancing handwritten character recognition.[20]

2.3 Summary linking literature review with the project:

Year Article/Author Tools/Softwa Technique Source Evaluation

re

Rohit Chandra Joshi; IEEE AI an

Genetic Signal Processing Accuracy an
Vivek Raj Patel; Anjali Python,R
Algorithm Conference effectiveness
Goyal RapidMiner
2022 evaluated

Year Article/Author Tools/Softwar Technique Source Evaluation

e

25
 International
Ashraf Y. Maghari,
CNN
Mahmoud M. Abu Conference on
Mining Advances in Accuracy an
Ghosh
Travel Electrical effectiveness
2017 Pattern Engineering evaluated

and

Computer
Applications

Ayush IEEE
HDR
Kumar Agrawal, A.K. Intelligen

Shrivas, Vineet kumar Machine Transportation Accuracy

2021 Awasthi Learning Systems evaluated
Conference

ACM
Python, SIGSPATIAL
n Mariam Omar
Sciki learn Workshop o
Alshrief,
Wafa El-Tarhouni, Computational Accuracy an
Machine Transportation effectiveness
Kenz Amhmed Bozed
2023 Learning Science evaluated

Machine Vilnius
CNN
Gadliauskas an Learning University Ope Accuracy
2022 Krisciunas Algorithm Series evaluated

Python, IEEE System

LTMSE
Man, an Accuracy an
Regression
Deep Cybernetics effectiveness
2020 Li, Lu, and Hao sklearn learning, Conference evaluated

26
 IEEE
Tiwari, Python, International
Chaturvedi, an Arya Advance Accuracy an
Matplotlib
Map Computing effectiveness
2021 Regression Matching Conference evaluated

Year Article/Author Tools/Software Technique Source Evaluation

Python,
Singh, Wu Travelling Accuracy an
Numpy, panda
Xiang, an Salesman IEEE Big effectiveness
2022 Krishnaswamy Regression Problem Dat Conference evaluated

Python IEEE
Support Intelligen Accuracy an
Regression
Sachin Chauhan, Vector Transportation effectiveness
2023 KNN Regression Systems evaluated
Samar Mahmood, T.
Conference
Poongodi

Python Urban Tri IEEE Servic

Priyadarshini,
Information Operations an Accuracy an
Satnam Singh, Anil Sklearn
Service Logistics effectiveness
Kumar Dixit
2019 LTMS Platform Conference evaluated

IEEE
Transactions o
Real-Time Intelligent
Pandas
He, Jiang, Lam and Travel Mod Transportation Accuracy
2019 Tang Matplotlib Detection Systems evaluated

27
 Park, Murphe IEEE

LTMS Transactions o
McGee, Kristinsson,
Intelligent Intelligent Accuracy an
Kuang,an Phillips Sklearn
Trip Transportation effectiveness
2014 Python Modeling Systems evaluated

Soares, Revoredo,
Baião, Quintell and
Python IEEE
Campos Random Accuracy
Transactions o
2019 Regression Forest evaluated
Intelligent

Year Article/Author Tools/Software Technique Source Evaluation

Transportation
Systems

Journal o
Global Researc
Python
Wadje an Khatri Genetic in Compute Accuracy
2021 SVM Algorithm Science evaluated

Traffic PARADISE
Seaborn
Tao, Sun, an Volume Research Accuracy
2019 Boukerche Pandas Prediction Laboratory evaluated

Sklearn
Accuracy an
Seaborn
Nafi, Khan, an Vehicular effectiveness
2014 Gregory NumPy Ad hoc RMIT evaluated
Universit
Networ

28
 MOE Ke Accuracy an
Matplotlib
Zhu, Song Machine Laboratory fo effectiveness
2015 , an Zhang Pandas Learning Transportation evaluated

Goole Maps
Mendapara, Krish
2016 Pabani, Yash Paneliya Seaborn Decision Tre Not specified Accuracy

Elizabeth Rani. G; MOE Ke

Sakthimohan. M; Pandas Laboratory fo
Abhigna Artificial Accuracy an
Reddy. G; NumPy Urban
Selvalakshmi. D; Neural effectiveness
Transportation
2009 Thomalika Keerthi; Matplotlib Network evaluated
Raja Complex
Sekar

29
Year Article/Author Tools/Software Technique Source Evaluation

Systems Theor
and Technolog

Support Accuracy an
Vector effectiveness
2004 Wu, Ho, and Le Various Regression Not specified evaluated

2.4 Objectives

The objective of the project "Handwritten Digit Recognition using Machine Learning" is to develop
a robust system capable of accurately identifying handwritten digits from images. Leveraging the
power of machine learning techniques, particularly Convolutional Neural Networks ( CNNs) and
Support Vector Machines (SVMs), the project aims to create a model that can efficiently handle the
complexities and variabilities present in handwritten digit data.

Handwritten digit recognition has been a fundamental problem in the field of pattern recognition and
machine learning. It finds applications in various domains including postal automation, bank cheque
processing, and digitizing historical documents. The objective of this project is to build a system that
can automatically recognize and classify handwritten digits into their respective numerical values (0 -
9) with high accuracy.

The project focuses on two primary machine learning techniques: Convolutional Neural Networks
(CNNs) and Support Vector Machines (SVMs). CNNs are highly effective for image classification
tasks due to their ability to automatically learn relevant features from input images. SVMs, on the
other hand, are powerful classifiers known for their ability to handle high-dimensional data and
provide robust classification boundaries.

30
Before training the models, the input handwritten digit images need to be preprocessed. This involves
tasks such as normalization, resizing, and noise removal to ensure uniformity and enhance the quality
of input data. Proper preprocessing steps are crucial for improving the performance of machine
learning models.

The project involves training separate models based on CNNs and SVMs using labeled datasets of
handwritten digit images. During training, the models will learn to extract relevant features from the
input images and classify them into one of the ten possible digit classes. Training involves
optimization of model parameters to minimize classification errors and improve overall accuracy.

Once trained, the performance of the models will be evaluated using a separate testing dataset.
Evaluation metrics such as accuracy, precision, recall, and F1-score will be used to assess the models'
performance and identify areas for improvement. Comparative analysis between CNNs and SVMs
will be conducted to determine which approach yields better results for handwritten digit recognition.

The final objective of the project is to deploy the trained models into a real-world application where
users can input handwritten digit images, and the system will provide accurate predictions. The
deployment phase involves integrating the models into a user-friendly interface and ensuring
scalability and efficiency in handling digit recognition tasks in real-time.

The project "Handwritten Digit Recognition using Machine Learning" aims to leverage the power of
CNNs and SVMs to develop an accurate and efficient system for recognizing handwritten digits. By
combining advanced machine learning techniques with effective data preprocessing and model
evaluation strategies, the project seeks to achieve high accuracy and robustness in digit recognition
tasks. Ultimately, the deployed system will find practical applications in various domains requiring
automated digit recognition capabilities, contributing to advancements in pattern recognition and
machine learning technologies.

31
2.5 Problem Solution

Handwritten digit recognition poses several challenges due to variations in writing styles, noise, and
distortions in images. Traditional methods often struggle to achieve high accuracy in digit
classification tasks. Therefore, there is a need for a robust solution that can accurately identify
handwritten digits from images, providing reliable results across different writing styles and quality
variations.

Solution:

1. Utilizing Convolutional Neural Networks (CNNs):

- Feature Learning: CNNs are adept at automatically learning relevant features from input images.
By employing CNN architectures, the model can extract hierarchical representations of handwritten
digits, enabling more effective classification.

- Spatial Hierarchy: CNNs leverage spatial hierarchy to capture local patterns and spatial
dependencies within handwritten digits. This spatial understanding helps in recognizing complex
patterns and variations present in handwritten characters.

- Data Augmentation: To enhance model generalization and robustness, data augmentation

techniques such as rotation, scaling, and translation can be applied. This helps in simulating variations
that may occur in handwritten digits, thereby improving the model's ability to classify diverse writing
styles.

2. Integrating Support Vector Machines (SVMs):

- Kernel Methods: SVMs are known for their effectiveness in high-dimensional spaces. By utilizing
kernel methods, SVMs can efficiently map input features to a higher-dimensional space, facilitating
nonlinear classification boundaries.

- Margin Maximization: SVMs aim to maximize the margin between different classes, leading to
robust classification boundaries. This helps in improving the model's generalization performance and
32
resilience to noise and outliers.

- One-vs-All Classification: In the context of multi-class classification, SVMs can employ the one-
vs-all strategy to train multiple binary classifiers, each distinguishing between one digit class and the
rest. This approach allows SVMs to handle multi-class classification tasks effectively.

3. Addressing Data Variability and Preprocessing:

- Normalization: Preprocessing techniques such as normalization ensure uniformity in input data

by scaling pixel values to a common range. Normalization helps in reducing the impact of variations
in brightness and contrast across different handwritten digit images.

- Noise Reduction: Techniques for noise reduction, such as Gaussian blurring or median filtering,
can be applied to remove unwanted artifacts or disturbances present in handwritten digit images. This
enhances the clarity of input images and improves model performance.

- Dimensionality Reduction: In scenarios with high-dimensional input data, dimensionality

reduction techniques like Principal Component Analysis (PCA) can be employed to reduce
computational complexity and enhance model training efficiency without sacri ficing classification
accuracy.

4. Model Evaluation and Optimization:

- Cross-Validation: To assess the performance of the models effectively, cross-validation

techniques such as k-fold cross-validation can be employed. This ensures robust evaluation by
partitioning the dataset into multiple subsets for training and testing.

- Hyperparameter Tuning: Optimization of model hyperparameters, such as learning rates,

regularization parameters, and kernel choices, plays a crucial role in improving model performance.
Techniques like grid search or random search can be used to identify optimal hyperparameter

33
configurations.

5. Deployment and Accessibility:

- User-Friendly Interface: The deployed solution should feature an intuitive user interface that
allows users to input handwritten digit images easily. A visually appealing and user-friendly interface
enhances accessibility and usability, making the solution more appealing to end-users.

- Real-Time Processing: Efficient implementation of the recognition system ensures real -time
processing of handwritten digit images, enabling quick and responsive results. Optimization
techniques such as model pruning and hardware acceleration can be employed to enhance inference
speed.

- Scalability and Adaptability: The solution should be designed to scale and adapt to varying
computational resources and deployment environments. This ensures flexibility in deployment across
different platforms and architectures, catering to diverse user requirements.

3. Design flow/Process

The design flow process for handwritten digit recognition begins with the collection and
preprocessing of a dataset comprising labeled handwritten digit images, such as the MNIST dataset.
During preprocessing, the images are normalized to a standardized pixel range and resized to a
consistent dimension to facilitate uniformity. Noise reduction techniques are applied to enhance
image quality, preparing the dataset for subsequent processing. Feature extraction is then carried out
using Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs). CNNs
hierarchically extract features from the images through convolutional layers, capturing patterns and
structures at different levels of abstraction. Conversely, SVMs represent images as feature vectors,
employing techniques like Histogram of Oriented Gradients (HOG) or pixel intensities. Following
feature extraction, the models undergo training using the preprocessed data. CNNs are trained using
backpropagation and optimization algorithms like Stochastic Gradient Descent (SGD), while SVM
34
classifiers are trained using the extracted features, with appropriate kernel functions and
regularization parameters selected through cross-validation. Subsequently, model evaluation is
conducted, assessing performance metrics such as accuracy, precision, recall, and F1-score on a
separate validation dataset. Cross-validation techniques are employed to gauge model generalization
and robustness. To further enhance performance, model optimization techniques such as
hyperparameter tuning and ensemble methods may be applied. Finally, the trained models are
deployed, integrating them into a user-friendly interface where users can input handwritten digit
images for real-time processing. This deployment phase ensures seamless integration, compatibility,
and efficiency of the recognition system, thereby offering practical utility in digit recognition
applications.

3.1 Concept Generation

In the domain of handwritten digit recognition, concept generation serves as the creative cornerstone
for devising novel methodologies aimed at enhancing the accuracy, efficiency, and adaptability of
recognition systems. One innovative concept involves the fusion of diverse machine learning
paradigms, particularly the amalgamation of Convolutional Neural Networks (CNNs) and Support
Vector Machines (SVMs) into a unified hybrid model. This hybridization strategy seeks to
synergistically leverage the unique strengths of each approach, such as CNNs' capability to
automatically learn hierarchical features from input images and SVMs' proficiency in constructing
robust decision boundaries. By integrating these complementary techniques, the hybrid model aims
to achieve superior performance in accurately classifying handwritten digits across various datasets
and writing styles.

Another promising avenue for advancing handwritten digit recognition lies in the realm of transfer
learning, a methodology that harnesses knowledge acquired from pre-trained deep learning models
to expedite learning on new tasks. In this context, pre-trained neural network architectures, such as
those trained on vast image repositories like ImageNet, can be repurposed for feature extraction in
handwritten digit recognition. By leveraging the rich representations learned from diverse image data,

35
these pre-trained models offer a foundation upon which to build specialized classifiers for digit
recognition tasks. Fine-tuning these models on handwritten digit datasets enables the adaptation of
high-level features to the nuances of digit classification, potentially enhancing both accuracy and
efficiency.

Furthermore, the augmentation and synthesis of training data present a compelling concept for
fortifying handwritten digit recognition systems against variability and data scarcity. Advanced data
augmentation techniques, including geometric transformations, image warping, and noise injection,
can be employed to generate diverse synthetic samples from existing data. By artificially expanding
the dataset's diversity and size, these techniques mitigate overfitting, enhance model generalization,
and improve resilience to variations in handwriting styles and image conditions. Moreover, generative
adversarial networks (GANs) offer a cutting-edge approach to data synthesis, enabling the creation
of realistic handwritten digit images that closely mimic the character istics of authentic data.
Integrating these synthesized samples into the training pipeline enriches the model's exposure to
diverse patterns and enhances its ability to generalize to unseen instances.

36
3.2 Evaluation & Selection of Specifications/Features:

In the evaluation and selection phase of specifications and features for handwritten digit recognition,
a comprehensive analysis is conducted to determine the most effective components for inclusion in
the recognition system. This phase is pivotal in ensuring that the chosen specifications and features
lead to a highly accurate and reliable recognition system. Firstly, key evaluation metrics such as
accuracy, precision, recall, and F1-score are defined to quantitatively assess the performance of the
recognition models. These metrics provide valuable insights into the models' ability to correctly
classify handwritten digits while minimizing false positives and false negatives.

To establish a performance baseline, baseline models are developed using conventional machine
learning algorithms such as logistic regression and decision trees. Through this process, various
features are explored, including raw pixel values, edge detection, and gradient orientations, to
determine their impact on model performance. This iterative experimentation allows for the
identification of feature sets that contribute most significantly to accurate digit recognition.

Furthermore, the evaluation extends to deep learning architectures, particularly Convolutional Neural
Networks (CNNs), which have shown remarkable efficacy in image classification tasks. Different
CNN architectures, ranging from classic models like LeNet-5 to more advanced ones like VGG and
ResNet, are tested and compared to ascertain their suitability for handwritten digit recognition.
Hyperparameter tuning techniques, such as adjusting learning rates and dropout rates, are employed
to optimize the performance of these models.

Transfer learning offers another avenue for enhancing recognition accuracy by leveraging pre-trained
deep learning models. By fine-tuning these models on handwritten digit datasets, researchers can
adapt high-level features learned from extensive image datasets to the nuances of digit recognition
tasks. Ensemble methods, including model ensembling and diverse model architectures, are also
explored to combine predictions from multiple models and further improve recognition performance.

37
3.3 Design Constraints:

Design constraints are pivotal factors that shape the development of a handwritten digit recognition
system, influencing its functionality, performance, and feasibility. Computational resources represent
a significant constraint, as the availability of computational power dictates the complexity of models
and dataset sizes that can be processed efficiently. Furthermore, memory and storage limitations
impact the scale of datasets and model parameters that can be accommodated during training and
inference, necessitating strategies such as data compression and efficient memory management to
address these constraints. Additionally, processing speed is a critical consideration, particularly in
applications requiring real-time or near-real-time inference, where models must be optimized for
swift execution. Techniques such as hardware acceleration and algorithmic optimizations are
employed to meet stringent performance targets without compromising accuracy.

Power consumption emerges as a key constraint in embedded or mobile applications, where energy
efficiency is paramount to prolong battery life and reduce power consumption. Models and algorithms

38
must be designed with a focus on minimizing computational complexity and optimizing energy usage
to mitigate these constraints effectively. Data availability and quality pose challenges in scenarios
with limited labeled training data or variability in data quality. To address these constraints, data
augmentation techniques and robust preprocessing methods are employed to enhance model
robustness and generalization capabilities across diverse datasets and conditions.

Latency and response time constraints are particularly relevant in applications where real -time or low-
latency processing is essential. Models and algorithms must be optimized to meet stringent latency
requirements while maintaining high accuracy levels. Security and privacy considerations impose
constraints on data handling, model deployment, and communication protocols, requiring robust
encryption and authentication mechanisms to safeguard data privacy and prevent unauthorized
access. Regulatory compliance with standards such as GDPR and industry-specific regulations further
constrains data handling and deployment practices, necessitating adherence to stringent compliance
requirements. Scalability and maintainability constraints drive the design of modular and extensible
architectures that can accommodate growing data volumes and user demands while facilitating ease
of maintenance and updates. By carefully navigating these design constraints, developers can
engineer handwritten digit recognition systems that balance performance objectives with practical
limitations and considerations, thereby ensuring their effectiveness and viability in real -world
applications.

39
3.4 Analysis and Feature Finalization

In the analysis and feature finalization phase of developing a handwritten digit recognition system, a
thorough examination of various features and their impact on model performance is undertaken. The
focus is on determining which features are most relevant and discriminatory in accurately
distinguishing between handwritten digits. Techniques such as correlation analysis and feature
importance ranking are employed to identify the most informative features for digit recognition.
Additionally, dimensionality reduction methods, including Principal Component Analysis (PCA), are
utilized to reduce the feature space's dimensionality while preserving essential information, thus
enhancing model efficiency and mitigating the curse of dimensionality.

The performance of the recognition models is rigorously evaluated using metrics such as accuracy,
precision, recall, and F1-score on validation and test datasets. Cross-validation techniques are applied
to estimate the models' generalization performance and detect overfitting. Comparative analysis is
conducted to assess the performance of different models, including conventional machine learning
algorithms, deep learning architectures, and ensemble methods. This allows for the identification of
the most effective approach for handwritten digit recognition, and benchmarking against existing
state-of-the-art methods and datasets helps gauge the system's competitiveness and effectiveness.

Once the analysis is complete, the final feature set is determined, selecting the optimal subset of
features based on their performance and computational efficiency. Feature scaling techniques, such
as normalization or standardization, are applied to ensure uniformity and facilitate convergence
during model training. The interpretability of the models is also considered, with visualization
techniques like t-SNE plots and explainable AI methods such as SHAP (SHapley Additive
exPlanations) used to understand the learned representations and explain model predictions.
Sensitivity analysis is conducted to evaluate the recognition system's robustness to variations in input
data, ensuring its reliability and effectiveness across different scenarios and conditions. Through this
thorough analysis and feature finalization process, the handwritten digit recognition system can
achieve optimal performance and accuracy in real-world applications.
40
3.5 Design Flow

The design flow for developing a handwritten digit recognition system involves a structured
progression through various stages aimed at creating an accurate and efficient recognition model.
Initially, the process begins with data collection and preprocessing, where a comprehensive dataset
of handwritten digit images is gathered and prepared. This includes tasks such as normalization,
resizing, and noise reduction to ensure uniformity and enhance image quality. Following data
preprocessing, feature extraction techniques are applied to extract relevant features from the digit
images. This step involves exploring different feature representations, such as raw pixel values or
more advanced feature descriptors like edge detection and gradient orientations, to capture the
discriminative characteristics of handwritten digits.

Once features are extracted, the next stage involves model selection and training. This includes
experimenting with different machine learning algorithms, such as Convolutional Neural Networks
(CNNs), Support Vector Machines (SVMs), or decision trees, to identify the most suitable approach
for the recognition task. Models are trained using the preprocessed features and labeled digit images,
with hyperparameter tuning and cross-validation techniques employed to optimize model
performance and ensure generalization to unseen data.

After training, the performance of the models is evaluated using validation datasets, assessing metrics
such as accuracy, precision, recall, and F1-score to gauge their effectiveness in digit recognition.
Comparative analysis may be conducted to compare the performance of different models and select
the best-performing approach. Additionally, model interpretability techniques, such as visualization
methods or feature importance ranking, may be employed to gain insights into the model's decision-
making process and identify key discriminative features.

Once a final model is selected, it undergoes deployment and integration into a real-world application

41
or system. This involves developing a user-friendly interface for users to input handwritten digit
images and receive accurate predictions. The deployed system is tested extensively to ensure
reliability, scalability, and efficiency in handling digit recognition tasks in real-time. Continuous
monitoring and refinement may be performed to improve the system's performance over time and
adapt to changing requirements or environments. Through this structured design flow, a robust and
effective handwritten digit recognition system can be developed to meet various application needs.

3.6 Design Selection

In the design selection phase of developing a handwritten digit recognition system, the focus is on
evaluating and selecting the most appropriate design approach based on various criteria, including
accuracy, efficiency, scalability, and practical feasibility. This phase involves careful consideration
of different design options and their potential impact on the system's performance and usability.
Several key steps are involved in the design selection process:

1. Evaluation of Design Alternatives: Various design alternatives, such as different machine learning
algorithms, network architectures, and feature extraction techniques, are evaluated based on their
ability to accurately classify handwritten digits. Each design option is assessed against predefined
criteria, including recognition accuracy, computational efficiency, and scalability.

2. Performance Analysis: The performance of each design alternative is analyzed using validation
datasets, assessing metrics such as accuracy, precision, recall, and F1-score. Comparative analysis
may be conducted to identify the design option that achieves the highest level of performance across
multiple evaluation metrics.

3. Efficiency Considerations: Design options are evaluated in terms of their computational efficiency,
including training and inference time, memory usage, and power consumption. Efficient design
choices that optimize resource utilization while maintaining high accuracy are prioritized.

42
4. Scalability and Generalization: The scalability of each design option is assessed to ensure that the
recognition system can handle increasing data volumes and adapt to diverse handwriting styles and
variations. Models that demonstrate robustness and generalization capabilities across different
datasets are preferred.

5. Feasibility and Implementation: Practical feasibility considerations, such as availability of

computational resources, data accessibility, and deployment constraints, are taken into account.
Design options that are feasible to implement within the given constraints are selected for further
development.

6. Stakeholder Input and Feedback: Input and feedback from stakeholders, including end-users,
domain experts, and project stakeholders, are considered in the design selection process. Stakeholder
preferences and requirements are taken into account to ensure that the chosen design option aligns
with their needs and expectations.

7. Risk Assessment: Potential risks associated with each design option, such as model complexity,
data scarcity, and deployment challenges, are identified and assessed. Mitigation strategies are
developed to address these risks and minimize their impact on the project.

By systematically evaluating and selecting the most appropriate design option based on these
considerations, a robust and effective handwritten digit recognition system can be developed to meet
the desired performance objectives and application requirements.

3.7 Implementation Plan

43
This code is a Python script designed to capture images of handwritten digits for a recognition
project. It utilizes the `time` module to introduce an 8-second delay between each image capture,
allowing the user time to draw a digit on the screen. The `ImageGrab` module from the Python
Imaging Library (PIL) is employed to capture screenshots within a specified region of the screen
defined by a bounding box. These images are saved sequentially to designated folders
(`captured_images/0/`, `captured_images/1/`, etc.), likely organized by the digit being drawn.
Each image is saved in PNG format with filenames corresponding to the iteration number,
prompting the user to clear the screen and redraw for subsequent captures. This process repeats
for a specified number of iterations, facilitating the collection of training data for the handwritten
digit recognition model.

44
 This Python script captures images of handwritten digits for a recognition project. It
employs the ` time` module to introduce a 5-second delay between each image capture, allowing
the user time to draw a digit on the screen. Utilizing the `ImageGrab` module from the Python Imaging
Library (PIL), it captures screenshots within a specified region defined by a bounding box. These
images are saved sequentially to designated folders (`captured_images/8/`, `captured_images/9/`,
etc.), likely organized by the digit being drawn. Each image is saved in PNG format with filenames
corresponding to the iteration number, prompting the user to clear the screen and redraw for
subsequent captures. This process repeats for a specified range of iterations, facilitating the collection
of training data for the handwritten digit recognition model.

45
 This Python script utilizes OpenCV (cv2) and CSV modules to process grayscale images of
handwritten digits captured for a recognition project. It iterates over each digit category, reading images
from respective folders and converting them to grayscale. Gaussian blur is applied to reduce noise, and the
images are resized to 28x28 pixels, suitable for input into a neural network. For each pixel in the resized
image, if its intensity is greater than 100, it's assigned a value of 1; otherwise, it's assigned 0. The pixel
values along with their corresponding labels are appended to a CSV file ('dataset.csv'), creating a dataset
formatted for training a digit recognition model.

46
 This code snippet utilizes Pandas to read a dataset from a CSV file ('dataset.csv') containing
pixel values of handwritten digits and their corresponding labels. The dataset is then separated into
independent variables (X), consisting of pixel values, and dependent variable (Y), representing digit
labels. Additionally, it installs required packages such as Matplotlib and Joblib. Finally, it uses Matplotlib
to preview a randomly selected image from the dataset along with its corresponding label.

47
 This code employs Joblib to load a pre-trained digit recognition model. It continuously captures screenshots
of a specified region of the screen, then preprocesses each image by converting it to grayscale, applying Gaussian
blur, and thresholding to binarize the image. After resizing the image to 28x28 pixels, it flattens the pixel values
and makes predictions using the loaded model. The predicted digit is then displayed on the screen alongside the
original image using OpenCV. The loop continues until the user presses the Enter key.

48
OUTPUTS:

49
4. Results analysis and validation
50
In the result analysis and validation phase of developing a handwritten digit recognition system, the
focus is on thoroughly evaluating the performance of the trained models and validating their
effectiveness in accurately classifying handwritten digits. This phase involves several critical steps
to assess the reliability, robustness, and generalization capability of the recognition system. Firstly,
the results obtained from the trained models are analyzed using various evaluation metrics, including
accuracy, precision, recall, and F1-score. These metrics provide quantitative measures of the models'
performance and help identify areas for improvement.

Furthermore, validation techniques such as cross-validation are employed to assess the models'
generalization performance and detect potential overfitting. By partitioning the dataset into multiple
subsets and training the models on different subsets while evaluating them on the remaining data,
cross-validation helps ensure that the models perform well on unseen data and are not overly sensitive
to fluctuations in the training dataset.

Comparative analysis may also be conducted to compare the performance of different models or
variations of the same model. This allows for the identification of the most effective approach and
helps inform decision-making regarding model selection and refinement strategies.

Additionally, qualitative analysis of model predictions and errors may be performed to gain insights
into the models' behavior and identify common patterns or challenges in digit recognition.
Visualization techniques, such as confusion matrices or heatmaps, can aid in understanding the
distribution of prediction errors and areas where the models struggle the most.

Validation of the recognition system also involves assessing its robustness to variations in
handwriting styles, image quality, and other environmental factors. Testing the models on diverse
datasets and real-world scenarios helps validate their reliability and effectiveness across different
conditions.

Finally, stakeholder validation and feedback play a crucial role in confirming the usability and
practical utility of the recognition system. Input from end-users, domain experts, and project
stakeholders helps ensure that the system meets their requirements and expectations, validating its
51
effectiveness in addressing the intended use cases and objectives. Through rigorous result analysis
and validation procedures, the handwritten digit recognition system can be thoroughly evaluated and
refined to achieve optimal performance and reliability in real-world applications.

3.8 Implementation of design

In our handwritten digit recognition project, the implementation of design is pivotal to achieving
accurate and efficient results. We start by preprocessing the dataset, which involves tasks such as
normalization to scale pixel values and augmenting the dataset for improved generalization. Next,
we select an appropriate model architecture. Convolutional Neural Networks (CNNs) are commonly
employed due to their ability to effectively capture spatial hierarchies in images. Our chosen
architecture consists of multiple convolutional layers followed by pooling layers to extract features
and reduce dimensionality.

Regularization techniques such as dropout are incorporated to prevent overfitting. We then train the
model using a suitable optimization algorithm like stochastic gradient descent (SGD) or Adam,
adjusting hyperparameters through experimentation to optimize performance. Additionally, we
employ techniques such as learning rate scheduling to fine-tune the training process.

Once trained, we evaluate the model's performance using metrics such as accuracy, precision, recall,
and F1 score on a separate validation set. We also perform extensive testing on unseen data to assess
generalization ability.

To further enhance performance, we may employ advanced techniques such as ensemble learning,
where multiple models are combined to make predictions, or transfer learning, leveraging pre-trained
models on larger datasets like ImageNet.

Finally, we deploy the model into a production environment, ensuring scalability and efficiency. This
involves considerations such as model serialization, containerization using tools like Docker, and
52
integration with web or mobile applications for real-world usage. Continuous monitoring and
updating of the model are essential to maintain its accuracy and relevance over time.

3.9 Final Preparation

In the final preparation stage of our handwritten digit recognition project, we focus on optimizing
every aspect for deployment and real-world usage. This involves meticulous attention to detail and
thorough testing to ensure the system's robustness and reliability. Firstly, we fine-tune
hyperparameters and model architecture based on performance evaluation results, aiming to achieve
the highest accuracy possible while maintaining computational efficiency.

We thoroughly validate the model on diverse datasets to ensure its generalization capability across
various scenarios and handwriting styles. Additionally, we optimize the inference pipeline for speed
and resource efficiency, considering factors such as batch processing and model quantization
techniques.

Furthermore, we implement comprehensive error handling and logging mechanisms to detect and
address any unforeseen issues that may arise during deployment. We also incorporate security
measures to safeguard sensitive data and prevent potential vulnerabilities.

Finally, we package the model into a deployable format, such as a RESTful API or a containerized
application, leveraging tools like TensorFlow Serving or Flask. Rigorous testing in staging
environments ensures smooth deployment into production, where the model can seamlessly integrate
with existing systems or applications.

53
 Throughout this final preparation phase, our focus remains on delivering a robust, accurate, and
scalable solution that meets the needs of our users while adhering to best practices in software
engineering and machine learning.

3.10 Project management, and communication

In our handwritten digit recognition project, meticulous project management and clear communication
are fundamental pillars driving its success. We adopt a structured approach, defining clear objectives,
timelines, and milestones right from the outset. Through collaborative tools and regular meetings, we
ensure that tasks are well-defined, assigned appropriately, and progress is tracked effectively.

Our communication strategy prioritizes transparency and inclusivity. We hold regular meetings to
discuss progress, address challenges, and align on priorities. Additionally, we maintain open channels
for asynchronous communication, enabling team members to share insights, raise concerns, and
collaborate seamlessly.

Furthermore, we foster a culture of accountability and empowerment within the team. Each member
understands their role and responsibilities, and we encourage proactive problem -solving and idea-
sharing. Feedback loops are integral, allowing us to continuously refine our approach and ensure that
we're delivering optimal results.

Externally, we maintain open lines of communication with stakeholders, providing regular updates on
project milestones, challenges, and successes. We actively seek feedback and input, ensuring that our
solution aligns with their expectations and requirements.

Overall, effective project management and communication are central to navigating the complexities
of our handwritten digit recognition project. By fostering collaboration, transparency, and
accountability, we strive to deliver a solution that exceeds expectations and delivers tangible value.

3.11 Testing /data validation

In our handwritten digit recognition project, thorough testing and data validation are crucial steps to

54
 ensure the reliability and accuracy of our model. We employ a multi-faceted approach to testing,
encompassing various stages of the project lifecycle.

Initially, during the data preprocessing phase, we rigorously validate the integrity and quality of our
dataset. This involves checking for missing values, outliers, and inconsistencies. We also perform
data augmentation techniques to enhance the diversity of our dataset, ensuring that our model
generalizes well to unseen data.

Once the preprocessing is complete, we partition our dataset into training, validation, and test sets.
The training set is used to train the model, while the validation set helps us tune hyperparameters and
monitor for overfitting. The test set serves as a final evaluation to assess the model's performance on
unseen data. During model training, we conduct extensive validation experiments to optimize
hyperparameters and architecture choices. We employ techniques such as cross-validation to ensure
robustness and prevent bias in our evaluations.

After training, we thoroughly evaluate the model's performance using various metrics such as
accuracy, precision, recall, and F1 score. We also analyze the model's confusion matrix to gain insights
into its strengths and weaknesses across different classes. Furthermore, we perform additional testing
on real-world data to assess the model's generalization capability and robustness. This may involve
collecting additional handwritten digit samples from diverse sources and environments.

Throughout the testing process, we maintain a systematic approach, documenting our experiments,
results, and any observations or insights gained. This allows us to iteratively improve our model and
ensure that it meets the desired performance criteria.

In summary, testing and data validation are integral components of our handwritten digit recognition
project, enabling us to build a reliable and accurate model that performs well in real -world scenarios.

3.12 Accuracy of Data

Ensuring the accuracy of the data used in our handwritten digit recognition project is paramount to
the reliability and effectiveness of our model. We employ several strategies to guarantee the quality
and integrity of the dataset.
55
Firstly, we meticulously source our data from reputable sources or generate it using controlled
processes to minimize errors and inconsistencies. This includes ensuring that the handwritten digit
samples are accurately labeled and annotated.

Secondly, we conduct thorough data preprocessing steps to clean and standardize the dataset. This
involves removing any corrupt or incomplete samples, addressing outliers, and normalizing pixel
values to a consistent range. By doing so, we mitigate the risk of introducing noise or biases into our
model.

Furthermore, we implement quality assurance measures throughout the data collection and labeling
process. This may involve manual inspection of samples by human annotators to verify accuracy and
correctness. Additionally, we utilize automated checks and validation techniques to detect and correct
any discrepancies or anomalies in the data.

During model training and evaluation, we continuously monitor the performance of our model on
validation and test datasets to assess its accuracy and generalization ability. This allows us to identify
any issues or inconsistencies in the data that may impact the model's performance and take corrective
actions as necessary.

Overall, by prioritizing data accuracy and implementing robust quality assurance measures, we
ensure that our handwritten digit recognition model is built on a solid foundation of reliable and
trustworthy data, ultimately leading to more accurate and dependable predictions.

4. Conclusion and future work

In conclusion, the development of handwritten digit recognition systems represents a significant
stride in the realm of artificial intelligence and pattern recognition, with profound implications across
various domains. These systems offer efficient and accurate means of interpreting handwritten digits,
facilitating tasks ranging from digitizing documents to automating data entry processes. Through

56
meticulous design, training, and validation, these systems have demonstrated remarkable capabilities
in accurately classifying handwritten digits with high precision. Looking ahead, the future scope for
handwritten digit recognition systems is promising, with several avenues for further advancement.
Future research endeavors may focus on enhancing recognition accuracy through the exploration of
advanced machine learning algorithms and deep learning architectures. Additionally, there is a
growing need for recognition systems capable of adapting to diverse handwriting styles and operating
in real-time on edge devices. Integration with emerging technologies such as augmented reality and
natural language processing could unlock new applications and use cases. Furthermore, the
integration of privacy-preserving techniques and the exploration of multimodal recognition
approaches offer opportunities for addressing privacy concerns and improving system versatility.
Overall, the future of handwritten digit recognition systems holds great potential for revolutionizing
data processing, enhancing efficiency, and enabling innovative applications across a wide range of
industries and domains.

4.1 Future Scope

Looking ahead, the future of handwritten digit recognition systems holds immense promise with
numerous avenues for advancement and innovation. Research endeavors are poised to focus on
enhancing both the accuracy and efficiency of recognition algorithms through the exploration of
advanced deep learning architectures, optimization techniques, and ensemble methods. Additionally,
there is a growing need for recognition systems capable of interpreting diverse handwriting styles,
including cursive writing and variations in stroke thickness, necessitating the development of
adaptable models trained on more diverse datasets. Moreover, the integration of multiple modalities
and the optimization of recognition algorithms for real-time deployment on edge devices are expected
to be key areas of focus. These advancements are anticipated to extend the application of handwritten
digit recognition systems into various domains such as healthcare and education, where they can
streamline processes, improve efficiency, and enhance decision-making. Furthermore, the integration
of privacy-preserving techniques and the exploration of synergies with emerging technologies like
augmented reality and natural language processing are poised to open up new avenues for innovation
and application. Overall, the future landscape of handwritten digit recognition systems is
characterized by ongoing evolution, driven by a quest for precision, efficiency, and adaptabil ity to
57
meet the diverse needs of users across industries and domains.

4.2 Conclusion

In conclusion, the development of a handwritten digit recognition system is a multifaceted process

that requires careful planning, implementation, and validation. By leveraging techniques such as data
preprocessing, feature extraction, and model training, it is possible to create accurate and reliable
models capable of classifying handwritten digits with high precision. Throughout the development
lifecycle, considerations such as computational efficiency, scalability, and robustness to variability
play pivotal roles in shaping the system's design and functionality. Through rigorous result analysis,
validation, and stakeholder feedback, the system can be iteratively refined to achieve optimal
performance and usability in real-world applications. As technology continues to advance, the
ongoing evolution of handwritten digit recognition systems holds immense promise for a wide range
of industries, including finance, healthcare, and education. With continued research and innovation,
these systems have the potential to revolutionize how we interact with handwritten information,
driving efficiencies, enhancing accuracy, and unlocking new opportunities for digital transformation.

4.3 Summary

In summary, the development of a handwritten digit recognition system involves several crucial
phases aimed at creating an accurate and reliable model. Beginning with data collection and
preprocessing, the system prepares a comprehensive dataset of handwritten digit images, ensuring
uniformity and enhancing image quality through normalization and noise reduction. Feature
extraction techniques are then employed to capture discriminative characteristics from the digit
images, followed by model selection and training. Throughout this process, careful evaluation of
design alternatives, performance analysis, and efficiency considerations guide the selection of the
most suitable approach. Result analysis and validation are conducted to thoroughly assess the trained
models' performance, utilizing evaluation metrics, cross-validation, and comparative analysis to
ensure reliability and generalization capability. Qualitative analysis of model predictions and
stakeholder feedback further inform refinements to the system. By systematically iterating through
58
these phases, the handwritten digit recognition system can achieve optimal performance, robustness,
and usability, meeting the diverse requirements and expectations of end-users and stakeholders in
various real-world applications.

4.4 References

1. Evaluation of Supervised Machine Learning Models for Handwritten Digit Recognition Rohit
Chandra Joshi; Vivek Raj Patel; Anjali Goyal 2022 4th International Conference on Advances
in Computing, Communication Control and Networking (ICAC3N)
10.1109/ICAC3N56670.2022.10074292

2. A Comparative Study on Handwriting Digit Recognition Using Neural Networks Ashraf Y.

Maghari, Mahmoud M. Abu Ghosh 2017 International Conference on Promising Electronic
Technologies.

3. A robust CNN model for handwritten digits recognition and classification, Saqib Ali, Zareen
Sakhawat, Tariq Mahmood, Muhammad Saqlain Aslam, Zeeshan Shaukat, Sana Sahiba 2020
IEEE International Conference on Advances in Electrical Engineering and Computer
Applications (AEECA).

4. A Deep Handwritten Digit Detection and Recognition Method Using a New Historical
Handwritten Digit DatasetHuseyin Kusetogullari, Amir Yavariabdi, Johan Hall , Niklas
Lavesson Big Data Research 2021 /10.1016/j.bdr.2020.100182

5. Handwritten Digit Recognition System Based on Convolutional Neural Network Jinze Li,
Gongbo Sun, Leiye Yi, Qian Cao, Fusen Liang, Yu Sun 2020 IEEE International Conference on
Advances in Electrical Engineering and Computer Applications (AEECA)

59
6. Handwritten Digit Recognition System Shubham Mendapara, Krish Pabani, Yash Paneliya
International Journal of Scientific Research in Computer Science, Engineering and Information
Technology doi.org/10.32628/IJSRCSEIT

7. MNIST Handwritten Digit Recognition using Machine Learning Elizabeth Rani. G;

Sakthimohan. M; Abhigna Reddy. G; Selvalakshmi. D; Thomalika Keerthi; Raja Sekar. R 2022
2nd International Conference on Advance Computing and Innovative Technologies in
Engineering (ICACITE) 10.1109/ICACITE53722.2022.9823806

8. Ensemble machine learning model for classification of handwritten digit recognition Mariam
Omar Alshrief, Wafa El-Tarhouni, Kenz Amhmed Bozed, 2022 IEEE 2nd International
Maghreb Meeting of the Conference on Sciences and Techniques of Automatic Control and
Computer Engineering (MI-STA) 10.1109/MISTA54861.2022.9837750.

9. A Robust Model for Handwritten Digit Recognition using Machine and Deep Learning
Technique Ayush Kumar Agrawal, A.K. Shrivas, Vineet kumar Awasthi 2021 2nd International
Conference for Emerging Technology (INCET) 10.1109/INCET51464.2021.9456118.

10. Recognition of Scanned Handwritten digits using Deep Learning Paras Nath Singh, Kiran Babu
T S 2023 IEEE 3rd Mysore Sub Section International Conference
(MysuruCon)10.1109/MysuruCon59703.2023.103969 20.

11. A Classical Approach to Handcrafted Feature Extraction Techniques for Bangla Handwritten
Digit Recognition Md. Ferdous Wahid, Md. Fahim Shahriar, Md. Shohanur Islam Sobuj 2021
International Conference on Electronics, Communications and Information Technology
(ICECIT) 10.1109/ICECIT54077.2021.9641406.

12. Improved Extreme Learning Machine Based on Deep Learning and Its Application in
60
 Handwritten Digits Recognition Xiao Xiao, Bolin Liao, Qiuqing Long, Yongjun He, Jun Li,
Luyang Han 2021 11th International Conference on Intelligent Control and Information
Processing (ICICIP) 10.1109/ICICIP53388.2021.9642213

13. MNSIT Handwritten Digit Recognition Using Machine Learning Classification Algorithms R
Anoop, Niharika Kiran, Bhargav P Raj, Ashwini Kodipalli, Trupthi Rao, B R Rohini 2023
World Conference on Communication & Computing (WCONF)
10.1109/WCONF58270.2023.10234977.

14. Algebraic Fusion of Multiple Classifiers for Handwritten Digits Recognition Huihuang Zhao,
Han Liu 2018 International Conference on Wavelet Analysis and Pattern Recognition
(ICWAPR) 10.1109/ICWAPR.2018.8521321

15. Sankhya: An Unbiased Benchmark for Bangla Handwritten Digits Recognition Aminul Islam,
Fuad Rahman, AKM Shahariar Azad Rabby 2019 IEEE International Conference on Big Data
(Big Data) 10.1109/BigData47090.2019.9005696.

16. Handwritten digit recognition based on classical machine learning methods Kairui Jin 2022 3rd
International Conference on Electronic Communication and Artificial Intelligence (IWECAI)
10.1109/IWECAI55315.2022.00040.

17. Handwritten Multi-Digit Recognition With Machine Learning Soha Boroojerdi, George
Rudolph 2022 Intermountain Engineering, Technology and Computing (IETC)
10.1109/IETC54973.2022.9796722.

18. Recognition of Handwritten Digit Using Different Machine Learning Algorithms Kumar
Priyadarshini, Satnam Singh, Anil Kumar Dixit 2022 International Conference on Cyber
Resilience (ICCR) 10.1109/ICCR56254.2022.9995767.
61
19. Handwritten Digit Recognition using Deep Neural Networks Sachin Chauhan, Samar
Mahmood, T. Poongodi 2023 4th International Conference on Intelligent Engineering and
Management (ICIEM) 10.1109/ICIEM59379.2023.10167391 .

20. An Empirical Comparison of Handwritten Character Recognition Using Machine Learning

Smitha N,Rahul Kumar Singh, Subodh Kumar Yadav, Sandeep Sah, Keshava Kumar N,
Hemanth Kumar H 2021 Second International Conference on Electronics and Sustainable
Communication Systems (ICESC) 10.1109/ICESC51422.2021.9532765

62