(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 13, No. 4, 2022

IoDEP: Towards an IoT-Data Analysis and Event

Processing Architecture for Business Process
Incident Management

Abir Ismaili-Alaoui Karim Baina Khalid Benali

Université de Lorraine, Rabat IT Center, ENSIAS, Université de Lorraine,
CNRS, Inria, LORIA, F-54000 Nancy, France Mohammed V University, CNRS, Inria, LORIA, F-54000 Nancy,
Rabat IT Center, Rabat, Morocco France
ENSIAS, Mohammed V University,
Rabat, Morocco

Abstract—IoT is becoming a hot spot area of technological Thus they don’t benefit from the different added values that
innovations and economic development promises for many in- could be created from sensor data for example, and the
dustries and services. This new paradigm shift affects all the useful knowledge that could be extracted from event logs and
enterprise architecture layers from infrastructure to business. historical data from previous executions. Further, BPM works
Business Process Management (BPM) is a field among others that in a reactive way [1] which is not sufficient when facing new
is affected by this new technology. To assist data and events ex-
plosion resulting, among others, from IoT, data analytic processes
radical or incremental changes. Early anticipation is crucial
combined with event processing techniques, examine large data to either avoid the occurrence of the problem or respond
sets to uncover hidden patterns, unknown correlations between to it quickly, with no latency, and in an efficient way. This
collected events, either at a very technical level (incident/anomaly lack of proactivity and predictability is remarkable in three
detection, predictive maintenance) or at business level (customer main steps of the BPM life cycle [2] [3]: the design and
preferences, market trends, revenue opportunities) to provide redesign step, the implementation step, and the execution step.
improved operational efficiency, better customer service and To address this problem, proactive-oriented concepts start to be
competitive advantages over rival organizations. In order to used in the BPM glossary such as proactive business process
capitalize the business value of data and events generated by management [1] [4], process forecasting or future-oriented
IoT sensors, IoT, Data Analytics and BPM need to meet in BPM [3] and context-aware business processes [5]. Therefore,
the middle. In this paper, we propose an end-to-end IoT-BPM
integration architecture (IoDEP: IoT-Data-Event-Process) for a
switching from a reactive to a proactive and adaptable business
proactive business process incident management. A case study is processes becomes mandatory for every organization. With this
presented and the obtained results from our experimentations new digitization of industrial processes, comes also the age
demonstrate the benefit of our approach and allowed us to of assistance, which mean that companies should be focused
confirm the efficiency of our assumptions. on customers, in order to offer a personalized and adaptable
services, and even predict their needs in almost real time. Deal-
Keywords—Business process management; internet of things;
machine learning; complex event processing; data analytics
ing with such a continuous changing environment requires an
intelligent, adaptive and flexible business processes. As a result
of all these new changes, organizations nowadays find that
I. I NTRODUCTION the traditional BPM systems present several limitations [6].
Nowadays, Business Process Management (BPM) is a well- In the literature, different approaches have bean proposed to
established discipline in both academia and industry. It is improve business process by applying diverse techniques and
considered as a powerful solution that helps organisations technologies such as recommended systems [7] [8], Ontologies
adapt to strategic, tactical and operational changes and gain [9], data analysis, data mining and process mining [10] [11],
more visibility and control over their business processes, complex event processing [1] [12], Ubiquitous Computing
so that they can continuously improve and optimise their [13], Internet of things - IoT [14] [15], just to name few.
activities and resources. Organizations use Business Process
Processes are executed within application systems belong-
Management systems as an activity-based workflow manager
ing to the real world, where humans, cooperative computer
that allow them to track the optimized functioning of their
systems and even physical objects are involved. In fact, con-
activities in order to gain in terms of agility, efficiency and
nected objects are becoming progressively more prominent in
performance. This method is mainly based on the concept
the business process execution environment. IoT represents the
of business processes. A business process is the structure of
inter-networking of physical objects [16] (also referred to as
activities and actions as they occur in the real world. It defines
”things”, ”connected devices”, ”smart devices”, ”ubiquitous
all the possible paths in the real process and the rules that
devices”), vehicles and other items embedded with sensors,
determine the path to follow and the actions to perform.
electronics, actuators, and network connectivity that enable
In most cases, business processes are isolated either from these ”things” to collect and exchange data when interacting
each other or from the organization’s external ecosystem. and sensing their environment. At the execution level, Busi-
www.ijacsa.thesai.org 900 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

ness processes can be classified into Person-to-Person (P2P), and evaluate our results. And finally, we conclude our paper
Person-to-Application (P2A), and Application-to-Application and present some future perspectives in Section 9.
(A2A) processes [2]. Recently with the emergence of these
IoT devices, new Business process interactions are emerging II. C ONTEXT AND M OTIVATING S CENARIO
such as Person-to-Thing, Thing-to-person, and Thing-to-Thing
due to the advent of Internet of things (IoT) technologies [17]. In this section, we present a real-life scenario to illustrate
However, IoT-BPM integration is still at its infant age. Most our problematic and highlight the challenges we are trying to
of the current research work on BPM-IoT integration, propose solve with the proposed approach.
new approaches that target a specific aspect of the BPM The case study of our research work belongs to silver
life-cycle. For example, updating business process models economy domain, which is a new industrial sector officially
by enriching business process model and notation (BPMN) launched in 2013 in France. The aim is to create personalized
with new elements that correspond to IoT domain and that services and new technologies to improve disability-free life
can explicitly define IoT devices within a business process expectancy and to help dependent elderly people as well as
[18], improving resource Optimization and monitoring and their care-givers on a day-to-day basis. Most countries all over
task execution via IoT context-specific knowledge provisioning the world live the demographic transition of aging population.
[19], improving business process execution via an IoT-aware According to the united nations, the number of people with
business process execution that exploits IoT for BPM by 80+ years old will triple between 2015 (126.4 millions) and
providing IoT data in a process-aware way [20], or proposing 2050 (446.6 millions). If we take France as an example, in
an architecture for IoT-BPM integration in order to cope with 2015 the number of people with more than 60 years old is
the issues and limitations raised by the recent case studies in 12 thousands, which represents 18% of the French population
both industry or academia. and they will represent more than 1/3 of the population by
In this paper, we focus on how to improve BPM through 20601 . Since the demographic change is becoming a global
IoT integration via an end-to-end architecture. This integra- phenomenon, several companies are focusing on developing
tion will help us to proactively manage the business process products and services to create age-friendly societies.
instances, that are launched by different IoT devices, based on The risk of diseases, loss of capacity and falls increase with
their priority level. Although academia and industry have taken age. Losing physical capacities due to age or some kind of
an interest in this integration, there is still a lot of research accidents can lead to serious falls of elderly people and those
work to be done in order to propose effective methodologies, falls can have adverse repercussions. In fact, The physical con-
design patterns and architectures to ensure efficient and smooth sequences of a fall differ from one individual to another. They
integration and communication between the IoT domain and can represent a decrease in mobility and an increase in daily
BPM. life activities dependency. Falls have also some psychological
In the literature, several research works are paving the way consequences such as a loss of self-confidence, which can
for BPM and IoT combination and integration, in order to accelerate functional capacities decline. Falls among seniors
optimize BPM using IoT and allow BPM to benefit from this result in a significant number of hospitalizations, with hip
new advanced technology. To go into more detail about our fractures being the main cause. Besides, falls are the leading
proposal related to IoT-BPM integration/Communication, it is cause of injury-related death.
appropriate to ask some research questions in order to define Several studies have been conducted in the field of silver
our problem in more concrete way. economy, in order to determine a standard definition of a fall
and the number of falls over a specific period to consider an
• Q01 : What is the state-of-the art regarding the inte- old person as a repetitive case. In [25] [21] [22] [23][24] a
gration/communication between IoT and BPM? fall represents “an unintentional change in position resulting
• Q02 : What design strategy or methodology can in coming to rest at a lower level or on the ground”. To
we follow in order to achieve a successful integra- characterize the repetitive aspect of a fall, we must determine
tion/communication between IoT and BPM? the number of falls and the time interval between falls. As
we can see in Table I, the majority of published studies
• Q03 : What are the encountered issues when integrat- consider at least two falls to retain repetitive character, with
ing IoT and BPM? an interval between two falls ranging from 6 to 12 months on
• Q04 : And given that data and event are the common average. Quick intervention after a fall, using a fall detector
points between IoT and BPM, to what extent can the for example, could avoid 26 % of hospitalizations, i.e. 160 M
functionalities offered by data analysis and complex Euros and 9,400 deaths per year. There are several solutions
event processing be exploited for an end-to-end IoT- for Fall Detection such as:
BPM architecture?
• Passive Solutions: where the senior must press a
The reminder of this paper is organized as follow. Section beeper to notify in case of an incident.
2 presents our context of work, illustrated with a real-life • Active solutions: these solutions require the use of
scenario. Section 3 overviews BPM, IoT and the integration sensors (Accelerometer, biological signals) or envi-
of these two technologies. In Section 4, we present in more ronmental detectors (presence, ground, doors...). In
details our approach. In Sections 5, 6 and 7 we present an case of a particular variation of the signals, the device
overview of our proposed IoDEP architecture from different triggers an alert.
perspectives. In Section 8 we describe the implementation of
our approach (via an initial block validation), and we discuss 1 https://www.insee.fr/fr/accueil

www.ijacsa.thesai.org 901 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

TABLE I. R EPETITIVENESS A SPECT OF A FALL

References Data Collected Participants Study plan Falls repetition
Questionnaire N = 730 Yes
[21] A follow-up period of one year Age >= 55 years Transversal >= 2/12 months
Postal questionnaire N = 1660 Yes
[22] A follow-up period of one year Age >= 70 years Transversal >= 2/12 months
Telephone questionnaire
Interval : 6 weeks N = 311 Yes
[23] A follow-up period of 36 weeks Age >= 70 years Observational >= 2/9 months
cohort
A follow-up period of 3 years
Participants report their falls weekly
on a fall calendar
Phone contact in case the person N = 1365 Yes
[24] is incapable of filling in his calendar Age > 65 years Observational >= 2/6 months
cohort
Participants interviews N = 377 Yes
[25] Retrospective (12 months) Age =78 +− 3 years Transversal >= 2/12 months

• Video-surveillance Solutions: the camera sensor an- they can detect hundreds of falls and risky situations and then
alyzes the senior’s behavior and triggers the alert assist the person in danger in less than five minutes. However,
accordingly. these ambient cameras videos generates a lot of false alerts,
triggered by an active person or a moving curtains for example,
Some of these solutions (bracelets, presence detectors, that are send to the back-office alert workflow management
active floors...) are ”blind”. This means that they do not allow system. Given that the human agent handles the received alerts
to know if a fall is a serious one or not, based on the received in a first in first out (FIFO) order, sometimes true and critical
alert. Only the image delivered by the video fall detectors, alerts may stay on a waiting list for few minutes until the
allows us to remove doubt about the incident and therefore agent handles all the false alerts or less critical ones that
avoid unnecessary interventions and therefore minimize the were received before the true one, as the incoming events are
overall cost of the service. Predicting and preventing falls intercepted, queued and launch the process instances . So if we
among elderly, is the main objective of our case study, in do not integrate a mechanism to help the human intelligence by
fact to apply our approach, we will use a data set and a prioritizing the event generated by a (very) serious case, high
business process model from a Video surveillance company. latency will induce delays that can be disruptive, depending
This company edits an automatic falls detection system for on the severity of the case.
elderly people and offers a 24/7 automatic alert solution and a
quick rescue without the intervention of the person in danger.
III. BACKGROUND AND L ITERATURE OVERVIEW
Fig. 1 depicts our incident management process, using
BPMN, which aims to manage falls alerts from detection to In the literature, several research works are focusing on
assistance and resolution. This incident management process is combining BPM and IoT. We start this section with a gen-
based on an analysis in real-time of alerts received from 24/7 eral BPM and IoT background. Then we present a literature
streaming cameras (IoT devices) for detecting elderly people’s Overview about BPM and IoT integration.
falls. Waiting too long (sometimes even for few minutes) can
be so risky as it can complicate the situation and also it can A. Introduction to Internet of things
be so painful for the person. That is why a quick rescue is
mandatory to assist the person after a fall or an incident. Smart objects swept in our life to facilitate it in so many
To achieve this prompt intervention, smart video surveillance ways and in different domains such as transportation, health
cameras are installed at client’s home or patients rooms at care, hospitalization, civil protection, smart home, smart cities,
geriatric services. These devices detect suspicious scenes that emergency, and individual automation. From smart phones
may be a fall or an incident, take a picture of the scene and then to new smart objects that interact not only with people but
send automatically an alert to the video-surveillance center. with other machines (Machine to machine communication).
This received alert is handled by a human agent, who qualify The concept of the ”Internet of Things” first emerged in
the alerts into four categories, as described below, and after that a presentation by K. Ashton, on the connection of Radio
he/she determines whether an assistance action is necessary or Frequency Identification (RFID) to the supply chain of Procter
not according to the criticality level of the alert. That is why, and Gamble in 1999. Since then, IoT has been exploding and
each alert received requires a quite vigilant treatment, in order invading our daily lives in different aspects (smart phones,
to be sure of its category, because the margin of error in this smart door locks, self-driving smart cars, smart cameras, smart
type of system must be very small, as those falls, in case of medical devices, etc.). This new technology started to thrive
a delayed intervention or an incorrect qualification, may have right after the development of the web in the 1990’s and mobile
an adverse impact on the person concerned: 1) False alerts Internet in the 2010’s. The internet of things (IoT) is growing
(level 0): Empty place. 2) False alerts (level 1): Active person. by leaps and bounds. It is made up of billions of smart devices
3) Alerts with average level (level 2): Seated person. 4) High that use wireless technology to communicate with each other
level alerts (level 3): Person lying down. and with us. IoT infrastructures can range from connected,
instrumented devices providing data to intelligent, standalone
Video-surveillance systems have proven their efficiency, as systems. IoT enables ”a world where things can automatically
www.ijacsa.thesai.org 902 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

Fig. 1. Motivating Scenario: Qualification and Assessment of the Risk Level of Incidents Process.

communicate to computers and each other providing services processes as a set of interrelated or interacting activities that
to the benefit of the human kind” [26]. transform inputs into outputs elements. For Harrington in [30],
he presents business processes as an activity or set of activities
Connected devices collect data by sensing their environ- that uses an input, adds some value to it, and delivers it as an
ment, and exchange data with other devices and humans. output to an internal or external customer. Dumas et.al [31]
All devices (things) act locally withing their environment. represent a business process as a collection of inter-related
Whereas, the IoT allows them to be remotely monitored via events, activities, and decision points that involve a number
existing network infrastructures, including the Internet [19]. of actors and objects, which collectively lead to an outcome
IoT contribute to continuously feeding ”big” data to every that is of value to a customer. All these definitions and many
node [27]. Besides, the evolution of the IoT invokes significant others present business processes as a set of activities and tasks
opportunities for private data exchange enabling new business that exploit the different resources (human and/or machine) of
models across heterogeneous networks [28]. However, it can- the organization to achieve one or more objectives previously
not generates value. That is why, it is necessary to couple it defined, in order to satisfy an internal or external customer.
with other technologies to transform this huge amount of data Each business process is attached to a single enterprise, but
into useful knowledge, in order to make meaningful decisions. in some cases it may interact with other business processes
belonging either to the same organization or to other organiza-
B. Introduction to Business Process Management tions. In order to achieve its objectives and ensure efficient per-
formance, the organization must subject its business processes
The Process approach has been increasingly adopted by to a continuous improvement mechanism. This mechanism
companies since the 1980s, leading to a new organizational represents the BPM life-cycle [31]. In fact, BPM life cycle is
model and a new way of operating in different organiza- about discovery, modeling/(re)desing, executing, analysis and
tions. Faced with a changing and competitive environment, monitoring of business processes in a perpetual repetition in
traditional approaches that treat the company as a closed order to optimize and automate these processes as much as
environment are no longer appropriate. Indeed, the process possible, and also to accommodate the ever-changing business
approach is a systemic approach that aims to transform the requirements.
hierarchical and vertical structure of an organization into a
transverse structure whose ultimate goal is the satisfaction of Business processes therefore occupy a very important place
external and internal customers. It is a method of analysis and in the field of information systems, because they play a core
modeling intended to ensure collaborative work in order to role within every organization. Moreover, the performance
control and improve the efficiency and smooth running of an level of any organization is indirectly linked to the efficiency of
organization. its processes and the quality of their models. In fact, a proper
management of business processes within an organization can
This method is mainly based on the concept of business have a very positive impact on the efficiency and smooth
processes. A business process is the structure of activities running of its activities, as it allows this organization to have
and actions as they occur in the real world. It defines all a clear vision of its objectives in order to better meet the
the possible paths in the real process and the rules that requirements of competitiveness that are constantly increasing.
determine the path to follow and the actions to perform
[29]. The International Standards Organization (ISO) defines BPM provides already different methods and solutions to
www.ijacsa.thesai.org 903 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

manage and analyse data and events. Among these methods we of event processing using the Complex Event Processing (CEP)
have Business Process Intelligence (BPI). In fact BPI systems engine. CEP is used to exploit and correlate large event streams
provides solutions to enhance decision making throughout a generated by heterogeneous data sources in order to produce
wide range of business activities, by analysing, predicting, useful information.
monitoring, controlling and optimizing business processes
[32]. Although, BPI has proven to be efficient for post- C. Event-Driven Business Process Management
execution prediction of future process behavior, it is unable
to manage and process huge amounts of real-time data and Although, IoT is becoming the hot spot area of techno-
events that are generates from different sources [33]. This logical innovations and economic development promises for
becomes more difficult when integrating IoT devices in a many industries and services, it still at its infant age, as we
BPM architecture. Another solution provided by the BPM have seen previously. During the last years, both academic
field, when dealing with real-time event data is Business and industrial world have been interested in this field and
activity monitoring (BAM). In fact, BAM is used in order to its integration with other domains. However, there is still
analyse data related to activities that have been executed. It a lot of research work to do to propose methodologies,
complements ex-post analysis of process execution by continu- design models and architectures in order to ensure an effi-
ously identifying specific situations at run-time and responding cient and smooth integration and bidirectional communication
to them by triggering specific actions [34]. However, this between IoT field and BPM. Dealing with event in BPM
technology remains less effective in use cases that includes field is not recent. Event-Driven Business Process Management
IoT generated events. The limitations of traditional BAM in (EDBPM) concept represents an enhancement of Business
IoT case studies can be seen from two aspects: prediction Process Management, by including other concepts such as
and proactivity. The first aspect is manifested in the complex SOA (Service Oriented Architecture), EDA (Event Driven
event correlation identification [35]. In fact, by sensing their Architecture), SaaS (Software as a Service), BAM (Business
environment, IoT devices generates a massive volume of event Activity Monitoring) and CEP (Complex Event Processing)
data that need to be processed and analyzed in order to extract [40]. This concept was first used in 2003 in a white paper of
useful information and to detect (complex) event patterns in Bruce Silver Associates in connection with the FileNet P8-
real-time. However, traditional BAM does not provide rule- BPM platform [41]. The idea behind EDBPM was limited to
based engines. This limitation becomes more apparent when a single event processing, because the concept of CEP was
the events are generated from diverse data sources, because not well known back then. So early applications of EDBPM
BAM lacks flexibility in integrating multiple heterogeneous was mainly focused on business process monitoring and on
data sources [33] [36] [37]. The second aspect is linked to the Key Performance Indicators (KPIs), or metrics measurement
absence of proactivity in BAM solutions. In fact, using BAM [42]. Despite the fact that the concept of CEP was used for the
in a reactive way is no longer sufficient, especially when we first time in 2002 by David Luckham is his book ”the power
have this huge amount of real life data and events. of events” [43]. It was until 2007 that an integration of CEP
in BPM field has been considered for the first time [44]. The
In business environment where every single event is im- integration of CEP technology within BPM is then known as
portant and need to be processed, Event Driven Architecture EDBPM.
(EDA) needs to be adopted. Now with the emergence of IoT,
events are becoming increasingly important for the current Recently, the EDBPM research area has been growing
information systems (SI), especially for organizations that significantly again due to the omnipresence of IoT devices.
integrate IoT devices and senors in their business operations We find applications that integrate IoT, CEP and (ED)BPM in
(video surveillance, Health care, ...). EDA is the successor of several research area and business sectors such as Health-care,
service Oriented Architecture (SOA). The idea behind this logistics, manufacturing, banking, smart cities/homes, cultural
paradigm is that everything is an event, all the different heritage, agriculture, etc. ([45], [46]). Hence the interest to
components of this architecture interact with each other by propose an architecture that integrates all these technologies
events. In this architecture each component is either an event side by side, in order to achieve a proactive event/instance
consumer or an event provider. Event consumers subscribe to management.
an intermediate event handler, and providers publish to that
handler. When the event handler receives an event from a D. Integration BPM with IoT: Literature Overview
providers, the handler forwards it to the consumer [38]. The During the last years, both academic and industrial world
difference between EDA and SOA is that SOA is based on have been interested in this field and its integration with other
the ”request/response” concept where the consumer of the fields. However, there is still a lot of research work to do
service sends a request to the producer, and the producer sends to propose methodologies, design models and architectures
a response that contains either the result or feedback. While in order to ensure an efficient and smooth integration and
EDA is based on the ”publish/subscribe” concept where the bidirectional communication between IoT and BPM.
communication pattern between the consumer and the provider
is reversed. So in this architecture consumers do not start the We discuss this related work from two perspectives. The
communication channel, but they receive the events published first perspective is the impact of IoT on business process.
by the event providers, which means that the communication The second one is the different problems and difficulties
is made in an unidirectional way [39]. encountered when we try to integrate IoT and BPM.
The main interest of the EDA is to manage real-time 1) How IoT Can Impact Business Process?: The alignment
process events and data in an efficient way. This message- of IoT and BPM is the focus of several research works.
driven architecture enables the introduction of a higher level However, this alignment comes with various challenges that
www.ijacsa.thesai.org 904 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

TABLE II. B USINESS P ROCESS R ELATED L IMITATIONS , A DDRESSED V IA I OT I NTEGRATION

BPM Life-cycle phases
BPM Limitations References
Process Process Process Process
modeling analysis execution monitoring
Physical surrounding X X [18] [47] [48]
Context-insensitivity X [47] [20] [49] [50] [51]
Data-input X [51] [16] [20] [47]
Models complexity X X [51]
Transparency X [51]
Latency X [51]
Event-logs quality X [51] [52]
Context-awareness X X X X [53] [54]
Knowledge feedback loop X X X X [55] [56]

need to be tackled. Both technologies will affect and of course Table (II) summarises and classifies these Business process
benefit from each other [19] [51]. However, in this paper, we related limitations according to the different phases of BPM
focus only on the influence / benefit of IoT on BPM. life-Cycle.
The integration of IoT can provide several benefits for 2) Problems Encountered in IoT/BPM Integration : The
BPM. Besides, it can also address some business process integration of IoT with BPM certainly contributes to business
related limitations. Among these limitations we have: process improvement. However, this integration comes with
• Physical surrounding:Business processes have no several issues and challenges. In [57], authors present the
access (or limited access) to physical surrounding, challenges that need to be addressed in Business Process Man-
as they operate in a cyber surrounding. Taking into agement Systems (BPMS) to achieve an efficient integration
consideration the physical surrounding at modeling of IoT, such as the absence of direct interaction between
phase can lead to correct resolution and execution of the business layer and the edge network, or the problem of
business processes [18] [47] [48]. complex and inflexible business process models due to a lack
of standardization and interoperability when modeling IoT
• Context-insensitivity: Business processes are insen- elements and components in BPM. The unexploited potential
sitive to context, they are considered as blind and of extracted data from sensing environment represents a chal-
stateless, which mean that in each business process lenging issue when integrating IoT with BPM [58]. Security
execution we do not take into consideration neither the and data privacy represent another level of IoT/BPM integra-
results from last process instances nor the context (a tion concerns [19]. Many other challenges exist and need to
context-aware business processes execution). So they be tackled when dealing with an IoT and BPM integration
need to know the conditions and situations in which and alignment, these challenges have been summarized by
IoT operates, given that IoT is by default context C.Janiesch et.al in their manifesto [19].
sensitive through devices/sensors [47] [20] [49] [50]
[51]. In this article, we focus mostly on the challenges linked to
event processing and also learning aspect withing an IoT-BPM
• Data-input: Business processes have no direct access architecture. The next section details this proposed approach.
to data generated by different devices ans sensors. This
data could be exploited to BP execution to progress
via taking actions (e.g., IoT-based trigger events/alerts, IV. P ROPOSED A PPROACH
IoT-based decisions...) [51] [16] [20] [47]. Event-driven business process management is mainly
• Models complexity: Integrating IoT technology can adapted in organizations that have a real-time based ac-
reduce the complexity of process models (for example, tivities which involved some sensors or some IoT devices
replace elements or patterns, ...). It Can also extend that collect data and generate new events by sensing their
and enrich process models. As a result, we have more environment [55]. However, a real-time system must have
precise process definitions that accurately reflect the three main characteristics to ensure better functioning inside
operational reality [51]. any organization [59]: 1) High availability, 2) Low latency
and 3) Horizontal scalability. Those three characteristics are
• Transparency: Integrating IoT technology enhances mandatory to achieve a real-time and efficient scheduling and
business process monitoring by increasing BP trans- event management in BPM.
parency through data provided by IoT sensors [51].
So in order to ensure a (near) real-time priority-based
• Latency: The incorporation of IoT technology can
business process instances management, we resort to an inte-
ultimately lead activity run time reduction and signif-
gration of four concepts: IoT (to sense the environment), CEP
icant latency that can result in an overall performance
(to detect situations of interest since it is considered as the
enhancement [51].
standard course for real-time analysis and situation detection
• Event-logs quality: IoT sensors produce a huge [60]), Machine Learning (to analyse our data, find patterns
amount of data that enrich process event logs. Given in it and then make predictions, to facilitate decision making),
that event logs are the fuel of process mining tech- and BPM (to manage our business processes). The idea behind
nology enriched event logs provide enriched process this integrated quaternity of technologies (see Fig. 2) is: 1) to
models [51] [52]. accompany data an event explosion resulting from IoT, 2) to
www.ijacsa.thesai.org 905 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

an end-to-end architecture aiming to provide a meet-in-the-

middle environment capable to capture data and event from
IoT sensors, when they are sensing their environment, create
actionable and useful knowledge, and allow this knowledge to
be used in the business layer through business processes.

B. Functional Requirements
Business processes are supposed to be smoothly executed
under different business situations and context. This constantly
changing environment, requires having business processes that
can easily be adapted to the appropriate action taken. However,
Fig. 2. Quaternity View of BPM, IoT, Machine Learning and CEP. without being coupled to other technology, business processes
are still deficient regarding the critical ability to provide
assistance to their users [62] due to a lack of two important
aspect:
examine large data sets in order to uncover hidden patterns,
unknown correlations between collected events, either at a very • Context-awareness: The emergence or even the om-
technical level (incident/anomaly detection, predictive main- nipresence of IoT solutions in different businesses
tenance) or at business level (customer preferences, market forces organizations to adapt their processes to a high
trends, revenue opportunities) 3) to capitalize business value level of connectivity. Context-awareness is a funda-
from data generated by IoT sensors, 4) to provide improved mental characteristic of ubiquitous computing [53],
operational efficiency, better customer service, competitive and it is the key to benefit from sensors collected raw
advantages over rival organizations. data, as it allows to store contextual information re-
lated to these raw data and to decide which data should
To turn this conceptual integrated quaternity of technolo- be processed, in order to facilitate the interpretation
gies into a concrete reality, we propose an end-to-end IoT-BPM [54] especially at the level of business processes.
architecture (IoDEP: IoT-Data-Event-Process). This integration
architecture follows Haze Architecture and Cascading Analyt- • Knowledge feedback loop: Traditional BPM systems
ics [61], incarnated by a DIKW (Data Information Knowledge present different limits, as they do not facilitate the
Wisdom) discovery pattern crossing the architecture from use of knowledge extracted/generated from data by
device then Fog/Edge to the cloud, and a learning feedback business processes after their execution. As a result,
loop that feeds forward insight to adjust either Fog/Edge tremendous amount of data and event data that are
or device algorithms [55]. IoT-BPM integration involves bi- constantly collected within the organization is not
directional communication. It is possible to acquire data/event exploited to improve business processes. As a matter
from sensors (e.g., monitor and control IoT devices) to manage of fact, these data represent for enterprises a real
business process instances and to send instructions to those engine of growth. However, a large amount of raw
devices (e.g., reset, adjustment or shut them down). One of data is not valuable; data must go through a whole
the requirement of our proposed architecture is scalability process to extract value from it [63]. The analysis of
without imposing an architecture redesign. That is the reason huge data helps organizations to extract information
behind using Haze Architecture and Cascading architecture as and then knowledge, because the real value is in
it ensures fluidity and dynamism. how organizations will use that data and turn their
organization into an information-centric company that
relies on insights derived from data analyses for their
A. Research Design / Modeling Methodology
decision-making.
As we said previously, in order to handle the challenges
faced by BPM when dealing with IoT objects, IoT and BPM To accompany data an events explosion resulting, among
need to meet in the middle. The Incorporation of two heavy others, from IoT, data analytic processes combined with event
paradigms such as IoT and BPM generates, with no doubt, processing techniques, examine large data sets to uncover hid-
a modeling methodology issue. This issue becomes more den patterns, unknown correlations between collected events,
complicated when this integration involves other technologies either at a very technical level (incident/anomaly detection,
(CEP and ML in this case). There are three types of modeling predictive maintenance) or at business level (customer pref-
methodologies: - The Top-down approach, - The Bottom-up erences, market trends, revenue opportunities) to provide im-
approach, - and the Meet-in the- middle approach. We propose proved operational efficiency, better customer service, compet-
in this article a meet-in-the-middle approach, to facilitate the itive advantages over rival organizations. In order to capitalize
integration of IoT and BPM. In fact, The meet in the middle business value of data and events generated by IoT sensors and
approach is considered as a method of refinement going alter- business process execution, IoT and BPM need to meet in the
nately from top to bottom to bottom to top. The combination middle, as we said previously. One critical use case for IoT
of deductive and inductive iterative sprints in this approach is to warn organizations when a product or service is at risk.
allows both re-use/mutualization and disruptive thinking. In Early detection is essential to either remedy the issue before
this paper, we propose an integration of IoT and BPM via it becomes a real problem or quickly do cleanup when failure
www.ijacsa.thesai.org 906 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

Fig. 3. ITIL Incident Management Process Functions vs IoDEP Architecture Analysis Matrix.

hits2 . Beside incident management process, IoT-BPM Architec-

ture integrates additional features to target highest IoT maturity
Supervision and incident management processes, especially levels. Among those features, we have:
in health-care domain, are considered as an event-driven busi-
ness process. The instances of these processes are: In some • Monitoring and Communication: All information,
cases, launched by IoT generated events (’Big’ data push metrics, and key performance indicators applicable to
paradigm: i.e. data is triggering processes). In order to handle the incident are assessed, recorded and reported (time
incident management in IoT-BPM architecture, Information spent on the incident, support actors, date and time
Technology and Infrastructure Library (ITIL) repository has of closure, number and type of reoccurring incidents,
been considered. In fact, according to ITIL V3 [64], incident average time to achieve incident resolution, percentage
management processes are composed of the following func- of incidents resolved at first line support that meet the
tions: Service Level Agreement, etc.).
• Prediction: Predicting an incident before happening,
• Detection / Identification: Detect the incident and will enable anticipatory incident management. This
identification of first elements of classification. may help avoiding the incident by actionning problem
management, or at least, this moves the predicted inci-
• Registration: Record basic details of the incident and
dent resolution closer to the incident detection insuring
propagate the incident alert as necessary.
proactive incident management. Thus, by collecting
• Categorization: Categorize incidents, Assign impact and combining connected devices/sensors data with
and urgency, and thereby define priority and match historical context data, IoT-BPM Architecture could
against known errors and problems. provide a wide variety of ad hoc, proactive and antic-
ipatory incident, anomaly, and problem management.
• Prioritization: The incident is prioritized for better
utilization of the resources and the Support Staff time Fig. 3 represents these functional requirements according
to each layer of our proposed architecture.
• Investigation and Diagnosis: Assess the incident
details, Collect and analyse all related information, V. OVERVIEW OF THE L AYERED A RCHITECTURE
and resolve, (including any work around) or route to This section provides a high level overview about the
online support. different layers of our proposed IoDEP architecture.
• Escalation: Escalate (functionally or hierarchically) Fig. 4 presents the layers of our architecture. In fact, the
where/when necessary. purpose of this layered architecture is to meet the requirements
detailed in the previous section. Our proposed architecture
• Resolution and recovery: Resolve the incident and requires five layers (see Fig. 4). IoT sensing layer or Edge
take recovery actions. Layer, IoT sensor data acquisition Layer, Detection, identifi-
• Closure: When the Incident has been resolved, the cation and registration Layer, Categorization and Prioritization
system should ensure that details of the action taken Layer, and Cloud layer. In the following we will present a
to resolve the incident are concise and readable, clas- detailed explanation of each layer of this architecture.
sification is complete and accurate according to root
A. The Edge Layer
cause, resolution/action is achieved.
This layer concerns all IoT sensors and devices, each
2 https://www.informationweek.com/big-data-analytics/when-internet-of- one of these have a precise role depending on their envi-
things-meets-big-data ronment, location and the purpose behind using them, such
www.ijacsa.thesai.org 907 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

previous steps.
• Pre-qualification: This step is reinforced with the
learning feed back loop [55], by learning from past
decisions, mainly false alerts. This final step will
consider and exploit the overall context to correct
future risk assessments.
At the end of these steps, the detected alerts/events are
sent to the cloud. Intercepted events are queued in order to be
qualified by a human resource.

B. The Fog Layer

When dealing with IoT devices and sensors, the cloud by
itself cannot connect, process and analyze data from thousands
and millions of objects and devices of different type and nature
spread out over vast areas. To overcome this issue, Fog layer
Fig. 4. The Architecture Layers. was introduced by Cisco in 2012, in order to offload the cloud
through the injection of smart devices at the network layer
to provide limited computational resources at the edge of the
device layer [68]. Fog computing and edge computing seem
as hospital, home, geriatric services, retail store, facilities. similar since they both consist of bringing intelligence and
IoT devices collect data and generate new events by sensing processing closer to data creation. However, the location of
their environment, after that those events could launch specific the intelligence, processing and computing power is the key
business process instances according to the particularities of difference between these two layers. Generally, intelligence
the detected situation. and computing power are placed in devices such as Smart
Cameras with embedded vision software (used in our case
As we have seen before, early detection is essential to either
study). While in the fog layer intelligence and computing
remedy the issue before it becomes a real problem or quickly
facilities are placed at the local area network (LAN).
do cleanup when failure hits [65]. The time lapse between
sensing the environment and sending a notification, a message In our proposed architecture, the fog layer is dedicated to
or a signal is very critical. In fact, latency is one of the most real-time stream processing with CEP as we can see in Fig. 6.
challenging requirements for connected IoT devices. One of
the most effective solution to reduce this lapse of time, is a) Complex Event Processing - CEP: An event is a
putting some computing on the device or at least bringing record of an activity in a system [69] and it represents any
computing near the device (edge computing). Pushing data change that occurs or will occur in this system, whereas a
to the edge eliminates long-distance data transmissions to the complex event is a set of events that are related to each other
cloud, which reduces network congestion and latency. Edge by aggregation, causality or time [70].
nodes are endowed with fall detection capabilities. In fact,
these nodes implements several advanced algorithms for image
processing and interpretation ([66], [67]). However, this is out
of scope of this paper. These algorithms allow the automatic
detection and qualification of risky events, following these
steps:
• Image Qualification: This step allows to qualify the
captured image of the camera. Several types of errors
are identified at this stage (Error of recovery of the
image, Image of incorrect size ...).
• Movement detection: This step allows to identify the Fig. 5. CEP Basic Architecture.
movements captured by the image, and then render
the behaviors of these movements. Mathematical fea-
tures are identified and extracted from the images Complex event processing is a widely used technology, it
and videos. Their exploitation allows to model the becomes an increasingly active research field [71] especially
behaviors and the events. with the popularity of publish/subscribe systems in 1997 [72]
• Movement tracking: This step consists of tracking as stated in [1]. Complex Event Processing (CEP) represents
objects in their attention zones. Tracking their evolu- a set of methods, techniques, and tools for processing events
tion allows to recognize an abnormal event or behavior while they occur [73]. It aims to process and analyze events
and to make a decision. generated from different sources in order to extract useful
information [74]. CEP is widely used to detect and deal with
• Decision: At this step a decision about launching an different business anomalies, threats and opportunities [74]
alert or not is taking, based on the results of the by analyzing event streams instead of traditional static data
www.ijacsa.thesai.org 908 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

stored in databases. CEP engines provide the scalability and the priority of the business process instances that need to be
the (near) real-time processing to filter, to combine and to executed.
extract actionable knowledge, known as situation of interest,
from a stream of events (see Fig. 5). C. The Cloud Layer
Complex situation of interest can be easily expressed using Both fog and edge computing are extensions of cloud
CEP engines and rules. For example, in our case study, queries networks. The majority of enterprises are already familiar with
such as: prioritize a case if the patient has some particular cloud since it is the de facto standard in most industries.
needs (Wheelchair, walker, etc.) or trigger an alert if the patient The concept of ”cloud” was used in several contexts in the
is a recidivist faller. 1990s, but only in 2006 when it became associated with
The combination between CEP and BPM is not recent. In offering services over the internet [82]. Cloud computing
fact, it has been widely used to control and monitor business is ”a model for enabling ubiquitous, convenient, on-demand
processes in real time in order to improve the effectiveness network access to a shared pool of configurable computing
of business operations by keeping track of what is happening resources (e.g., networks, servers, storage, applications, and
now and raising awareness of issues as soon as they can be services) that can be rapidly provisioned and released with
detected [75]. CEP helps to monitor not only process instances minimal management effort or service provider interaction.”
and activities during run time, but also different events that are [83]. Cloud layer facilitates storing and accessing data and
related to business process but not necessarily generated from programs over the internet (as a service) rather than on servers
the process instances [76]. Several approaches and solutions of the enterprise as we can see in Fig. 6. In fact, it offers
based on the integration of CEP with BPM have been proposed the ability to drastically outgrow an organization’s normally
for run time or design time ([77], [78], [79]) just to name a available storage, without having to host any additional servers.
few. In our IoDEP architecture, Data and intelligence are pushed
through layers from edge to cloud in order to be analysed
The starting point of this incident management approach and processed. The knowledge produced goes through a learn-
is the real time analysis of the incoming generated events, ing feedback-loop that feeds forward insight to adjust either
using CEP engine to detect the events with the highest priority Fog/Edge or device algorithms. To produce this knowledge, we
withing all the incoming ones, that will launch our business use machine learning algorithm. In the following, we introduce
process different instances. And based on this estimated pri- this technology and present in more details the algorithm used
ority we can schedule those instances. In fact, the business in our approach.
process instances triggered by those events will then have
the highest priority to be assigned to the available human a) Machine Learning: Integrating IoT in a BPM envi-
resources. In order to process the incoming events and then ronment generates several challenges, among these challenges,
detect meaningful patterns concerning the important situation we have the huge amount of data and event data that are
of interest (which represent for us the events with the highest continuously gathered. Data and event data are the key to get a
priority), CEP engine needs a set of rules that are determined in better understanding of the functioning of business processes.
advance. A rule for us in this approach represents a condition This data represent for enterprises a real engine of growth.
(IF...THEN ...) that characterizes the event source (the different However, a large amount of raw data is not valuable; data
sources that generate each event). So in the processing step, must go through a whole process to extract value from it. In
whenever a condition is satisfied by the event, the priority of fact, pre-processing and exploring data before using it help to
this event increases. get correct assumptions and insights in order to make correct
predictions and finally take correct and accurate actions and
CEP strength relies basically on concept of rules and decisions such as instances scheduling, resources management,
operators. CEP engines are mainly based on a set of rules or business process models redesign. The analysis of this huge
provided by a rule engine. A rule engine represents a part generated data helps organizations to extract information and
of a CEP engine that generates rule models. These rules then knowledge, because the real value is in how organiza-
are used in order to create and/or modify business logic in tions will use that data and turn their organization into an
a Business Process Management System (BPMS) [80] for ”information-centric company that relies on insights derived
example. CEP rules are based on CEP operators. Among from data analyses for their decision-making”4 . Hence inte-
these operators we have: aggregation operators, sequencing grating data analysis techniques, process mining, data mining,
operators, logic operators, single-item operators, windowing machine learning algorithms or even deep learning in each step
operators, and flow management operators. Those rules are, in of business process life cycle is very crucial for the process
most of the time, manually predefined by domain experts, and improvement. In our approach, we have chosen to exploit the
after that implemented in CEP systems such as Esper, Siddhi, machine learning algorithms in order to enhance one aspect of
FlinkCEP3 , or Oracle. Since defining these rules manually can business process life-cycle, which is the instance management.
be error prone and time consuming, there are many recent
approaches that propose an automatic CEP rules learning and Machine learning is a branch of the artificial intelligence
generation [1] [81] [74]. However, the automatic CEP rules research domain. Using mathematical methods, Machine learn-
learning and generation is out of the scope of this paper, ing enables systems to learn from data and generate knowledge
because the main purpose of this paper it to pave the way from experience. With time and more experience, the system
for the use of CEP in order to achieve a real-time analysis of can learn and improve and sharpen a model that can be
incoming events in order to detect situations of interest, about used to predict outcomes of questions using previous learning
3 https://nightlies.apache.org/flink/flink-docs-release-1.14/docs/libs/cep/ 4 https://datafloq.com/read/3vs-sufficient-describe-big-data/

www.ijacsa.thesai.org 909 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

[84]. Machine learning algorithms are organized into different VI. OVERVIEW OF THE F RONT- END AND BACK - END OF
categories based on the learning type: supervised learning, un- THE A RCHITECTURE
supervised learning, semi-supervised learning, Reinforcement
learning, and Transduction learning. Supervised learning and From a front-end and back-end perspective, our architec-
unsupervised learning algorithms are the most used in real ture can be seen as follows (see Fig. 3):
world problems. They can be further grouped into different The IoDEP Front-end architecture is composed of two
categories that we can find in real world machine learning parts:
problems. However, the choice of the appropriate algorithm
depends on your case study, the type of data manipulated, and • Connected Device/Sensor Processing and Analyt-
the purpose of your analysis. ics: insure acquisition of incident data, and incident
data filtering and Simple classifier.
The first step of our approach aims to ensure an efficient • Fog/Edge Processing and Analytics: insure the fol-
acquisition, filtering, and analyzing incident/event data gen- lowing functionalities:
erated by IoT devices. The second step consists on scoring 1) incident data processing: Detection of
the generated events from IoT devices, based on the result of anomaly incident.
the first step, using unsupervised machine learning algorithm. 2) incident data analytics: Pattern recogni-
We opted for clustering algorithms to discover groups in our tion/correlation/scoring (advanced supervised
dataset, in order to achieve a categorization of the event time-based analysis algorithms here need
sources that trigger our process instances. We choose K-means smaller training set but may need more per-
clustering algorithms, which is a partitioning technique used formance resources like GPU).
to analyse data based on the distance between different data 3) incident data routing: Transmission of
points in the input dataset. This algorithm was described by the anomaly information through an Edge
Hartigan in 1975. The idea behind K-means algorithm is to Spooler.
divide a dataset composed of M data points in N dimensions
into K clusters, in such a way that the within-clusters sum The IoDEP Architecture Back-End is represented by sev-
of squares is minimized [85]. The most complicated part of eral components insuring cloud side processing and analytic
this algorithm is the determination of the right value of K of incident management:
which represents the number of clusters. In the literature,
we can find several methods for selecting the most optimal • Cloud Data Processing and Analytics: Ensure the
number of clusters for this algorithm such as: The Elbow following functionalities:
method [86]: The Average Silhouette method [87], or the GAP 1) incident data routing: Transmission of the
statistic method [88]. When applying K-means algorithm in anomaly information to the relevant back end
our approach, we have used the Elbow method to determine processing and analytics system – ESB/CEP.
the value of K. The basic steps of K-means algorithm are 2) incident data analytics and intelligence:
shown in the following pseudo-code: Extraction, cleaning and annotation, In-
tegration, aggregation and representation,
Modelling and analysis Pattern recogni-
Algorithm 1 K-Means Clustering Algorithm tion/correlation/scoring (more sophisticated
Input: supervised machine learning algorithms (e.g.
S = s1 , s2 ..... sm // list of data points (list of sources which deep learning) may here need big training
generate the different events) sets) (Big Data).
K //Number of clusters 3) incident data processing: Anomaly Hu-
1: choose K Random data points from S as initial clusters man Processes (Human qualification of the
centroids anomaly information) and Enterprise Business
2: repeat Processes (BPMS based on Big Data analyt-
3: Assign each data point si to the cluster which has the ics), and Interpretation: Through on Report-
closest centroids. ing incident KPI Scoreboards based on (Data
4: Calculate the new centroids of each cluster. Warehousing and Data visualisation).
5: until Convergence //no more changes for centroids
Output: Set of K clusters VII. A PPLICATION A RCHITECTURE OVERVIEW
In this section, we present a concrete application of our
architecture (see Fig. 6). The content of the Edge layer can
differ from one case study to another. For our case study, IoT
We tested several criteria such as the frequency of falls or devices were represented by smart cameras that detect the falls
total number of falls, in order to have the most representative of patients. In other context, these devices and sensors could be
clustering for our data. We apply K-means algorithm on a set either Smart locks, Fire and smoke alarms, or smart wristband,
of events sources in order to classify those sources on different etc.
clusters using a score that we calculate for each event’s source
(a patient in our case) based on the frequency of previously Our Fog layer represents the real-time stream processing
generated events and their criticality value given previously by layer using CEP technologies. We propose the use of Apache
the agents (human resources) in the qualification step. Flink framework as it facilitates complex event processing
www.ijacsa.thesai.org 910 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

We worked with a dataset of patients falls (from 01-02-

2016 to 12-06-2017), this dataset contains 238228 observations
generated by 81 patients.The historical data in this dataset are
gathered from our incident management process past instances,
and they are partitioned as follows: 89312 alerts are of level 0
(low), 148466 of level 1(average), 275 of level 2 (serious) and
175 of level 3 (very serious). Since that the serious and very
serious alerts are the most important in our study, we have
applied our clustering algorithms on these two levels only.
The dataset was stored in a PostGreSQL database and
all analyses and algorithms implementation were conducted
using R and different R packages6 , such as: (tidyverse, RPost-
greSQL, ggplot2, dplyr, caret, ...). And All our experiments
were conducted on an Intel(R) Core(TM) i5- 540M2.53GHz.
All data have been anonymized.
Experimental results for K-means:
As we have seen before, to determine the value of K, which
represents the number of clusters, we have used the Elbow
method (see Fig. 7).

Fig. 6. IoDEP Applicative Architecture.

for real time analysis. Our choice is based on the bench-

mark that we have conducted. The Enterprise Service Bus
layer represents a message broker. It is an integration so-
lution implementing a totally distributed architecture where
the applications or services to be integrated are distributed
on different machines or information systems, and its role is Fig. 7. Elbow Method - Sum of Squared Distances.
to ensure communication and interoperability between these
different applications whatever their communication protocols.
ESB is mostly adapted to asynchronous communications, The plot represents the variation of the sum of squares
publish/subscribe messaging, and message queues. Solutions with the number of classes. We notice that the most adequate
such as Mule ESB, PEtALS, JBoss ESB, Glassfish ESB, or number of clusters would be 4, since this point corresponds to
Apache Camel, could be used in this architecture. Historical an inflection point and the sum of squares seems to stabilize
data in our architecture are stored in a PostgreSQL database. from this point on.
However other Data Base Management System (DBMS) can
be used. For the Business Process Layer, several Business As we can see in Fig. 8, patients used in this clustering
Process management Systems can be used such as Bonitasoft5 are divided into 4 clusters: patients with fewer than 8 falls
(that we have used in our case), Signavio, or Camunda. As (the blue cluster), 9 to 12 (the green cluster), 13 to 18 (the red
an application server, solutions such as Apache Tomcat, or cluster) and more than 24 falls (the black cluster). These latter
WildFly can be used in this architecture. As a web server represent the most critical cases. As we have seen before, The
to distribute web content, examples such as Apache web clustering algorithm helps us to categorize our event sources.
Server, can be integrated in this architecture. And finally, In the following we will present some experimental results
for communicating results a BI Dashboard layer, interactive that proves the interest of integrating CEP into the IoDEP
Dashboards can be build using PHP, or Shiny R package given architecture, in order to manage the event that will trigger the
that we have used R language in our experiments. process instances.
Experimental results for CEP integration:
VIII. E XPERIMENTS AND R ESULTS The purpose of this series of experiments is to show the
interest of integrating CEP in the IoT-BPM architecture. So
We opted for à block validation in order to conduct an we will compare two solutions. Solution 1 represents incident
initial validation of the efficiency of our IoEDP architecture. management in our architecture without the integration of
Experimental settings: CEP, and Solution 2 represents the integration of CEP in the
architecture.
5 https://fr.bonitasoft.com/ 6 https://www.tidyverse.org/

www.ijacsa.thesai.org 911 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

The different modules of this architecture operate as fol-

lows:
• Event Producer: Events in this approach are mostly
generated by some sensors or some IoT devices by
sensing their environment (Smart cameras).
• Lightweight Message Broker: To manage the
amounts of events received and that need to be pro-
cessed by the CEP engine, we use a message broker
that ensure the communication between the source and
the target based on a publish/subscribe mechanism.
This asynchronous mechanism implemented by mes-
sage brokers allows source and target messages to be
completely decoupled. Besides the message brokers
can as well store the messages locally until they can
Fig. 8. Clustering with K-means. be processed by the target element. That is why we
have chosen RabbitMQ.
• CEP engine: Flink CEP is used in this case to filter
CEP Rules: and process incoming events based on the predefined
rules, in order to detect the events with highest priority
Based on the context of our case study, we have defined among the incoming stream of events.
some rules that help us to estimate the priority level of the
incoming event. To define those rules we take into consider- • Event consumer: It represents in this approach a
ation the available information about the patient, his/her past business process management system (BPMS) where
incidents (falls) and the cluster to which this patient (event the processes are managed, executed and monitored.
source) belongs. Taking into consideration the results of our
clustering (detailed in the beginning of this section), we tried In our experiment for both solutions, we have simulated
to manually define some rules. So basically we come up with several event streams with different total number of events
the following rules: (100, 200, 300, 400, 500) (generated from the historical events
in our data set). Our objective is to compare the evolution
• IF the event source belongs to cluster 4 (the critical
of the computation time of both solutions (as described in
cluster) THEN the new event generated by this source
the following tables) as a response to increasing the number
might be serious.
of input events (non-concurrent access (see Table III and
• IF The patient has some particular needs (Wheelchair, Table IV) and concurrent access (see Table V and Table VI)).
walker, etc.) THEN the new event generated by this Besides, in both solutions we have used the results of the
source might be serious. clustering step.
• IF The last event generated by this patient within one Non-Concurrent Access (NCA):
month was a serious or very serious alert THEN the
new event generated by this source might be serious. TABLE III. C OMPUTATION T IME ( SEC ) FOR S OLUTION 1 - NCA

CEP Message broker: Total event number Computation time (sec) - Solution 1
100 19.0
200 19.45
As we have seen before, in order to manage the different 300 25.48
incoming events within a CEP solution, we need a message 400 31.3
broker. In this experiment, we have chosen RabbitMQ7 . It is an 500 37.4
open source message broker, lightweight and easy to deploy.
TABLE IV. C OMPUTATION T IME ( SEC ) FOR S OLUTION 2 - NCA
Global Schema of Event-pattern detection with CEP Total event number Computation time (sec) - Solution 2
solution: 100 22.07
200 32.14
Fig. 9 illustrates our first attempt to integrate CEP engine 300 44.25
in our IoT-BPM architecture. 400 55.6
500 70.62

Concurrent Access (CA) :

As we can see in Fig. 10, solution 1 presents better results
comparing to solution 2 when we have a non-concurrent access
Fig. 9. Priority-based Event Management with CEP. of the incoming events. However, when we have a concurrent
access, the CEP-based approach (solution 2) presents better
results, especially when we increase the number of incoming
7 https://www.rabbitmq.com/ events (see Fig. 11).
www.ijacsa.thesai.org 912 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

TABLE V. C OMPUTATION T IME ( SEC ) FOR S OLUTION 1 - CA

Total event number Computation time (sec) - Solution 1
200 48
400 72.1
600 108

TABLE VI. C OMPUTATION T IME ( SEC ) FOR S OLUTION 2 - CA

Total event number Computation time (sec) - Solution 2
200 27
400 34
600 60.3

Fig. 10. Input Events with Non-concurrent Access.

Although solution 1 seems to be more efficient at low
input events volumes, the CEP solution can performs better
especially if implemented in an effective IoT architecture with
’Big’ Data requirements.
For incident management systems providing a balance
between (near) real-time event processing and scalability, it is
very important to achieve an efficient and optimized business
process instances scheduling and event management in BPM.
Moreover, in real cases, we deal in most of the time with
concurrent access of incoming events. So this confirm the
efficiency of our assumptions that CEP can provide better
results when integrated to an IoT-BPM architecture, and it can
also provide better results comparing to traditional approaches
for business process instances scheduling. Those initial results
are encouraging to implement the entire end-to-end IoDEP Fig. 11. Input Events with Concurrent Access.
architecture.
Interested readers can check the complete solution that we
have implemented from GitHub8 . Fully implementing this architecture, extending to test more
machine learning algorithms, and exploring the automatic CEP
IX. C ONCLUSIONS AND P ERSPECTIVES rules learning in order to enrich the business rules engine for
proactive event processing and process instances execution, are
Integrating IoT and BPM as a step towards an improved all plans to realize on our future schedule.
process management that benefits from data and event power,
is possible via an integrated architecture (IoDEP). The idea
behind this architecture is to manage data and events at the R EFERENCES
same time via an integration approach that includes four [1] R. Mousheimish, Combining the Internet of things, complex event
concepts: IoT (to sense the environment), CEP (to detect processing, and time series classification for a proactive business
situations of interest since it is considered as the standard process management. PhD thesis, Université Paris-Saclay, 2017.
course for real-time analysis and situation detection), Machine [2] W. M. Van der Aalst, “Business process management: a comprehensive
survey,” International Scholarly Research Notices, vol. 2013, 2013.
Learning (to analyse our data, find patterns in it and then
make predictions, to facilitate decision making), and BPM (to [3] R. Poll, A. Polyvyanyy, M. Rosemann, M. Röglinger, and L. Rupprecht,
“Process forecasting: Towards proactive business process management,”
manage our business processes instances. in International Conference on Business Process Management, pp. 496–
512, Springer, 2018.
We introduced in this article the reason behind using
[4] A. Ismaili-Alaoui, K. Benali, K. Baı̈na, and J. Baı̈na, “Business process
machine learning algorithms and CEP techniques in our IoT- instances scheduling with human resources based on event priority
BPM communication approach. In fact, this bi-directional determination,” in International Conference on Big Data, Cloud and
communication is established through event in one direction Applications, pp. 118–130, Springer, 2018.
and data in another direction. That is why conducting a data [5] M. Anastassiu, F. M. Santoro, J. Recker, and M. Rosemann, “The
analysis approach with event management can facilitate this quest for organizational flexibility: driving changes in business pro-
communication/integration. Moreover, we have presented the cesses through the identification of relevant context,” Business Process
Management Journal, vol. 22, no. 4, pp. 763–790, 2016.
different functionalities proposed by this architecture and also
the different requirements that should be addressed. Through- [6] F. Bergenti, G. Caire, and D. Gotta, “Interactive workflows with wade,”
in 2012 IEEE 21st International Workshop on Enabling Technologies:
out the different overviews of our proposed architecture, we Infrastructure for Collaborative Enterprises, pp. 10–15, IEEE, 2012.
argue that our approach is generic and can be used in mul- [7] I. Barba, B. Weber, C. Del Valle, and A. Jiménez-Ramı́rez, “User
tivariate settings, and most importantly in normal and strict recommendations for the optimized execution of business processes,”
environments where time and priority matter. Data & knowledge engineering, vol. 86, pp. 61–84, 2013.
[8] M. A. Setiawan, S. Sadiq, and R. Kirkman, “Facilitating business
8 https://github.com/Abir-IA/CEPFlink EventManagement process improvement through personalized recommendation,” in Inter-

www.ijacsa.thesai.org 913 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

national Conference on Business Information Systems, pp. 136–147, [27] J. Rifkin, The zero marginal cost society: The internet of things, the
Springer, 2011. collaborative commons, and the eclipse of capitalism. St. Martin’s
[9] M. Khanbabaei, F. M. Sobhani, M. Alborzi, and R. Radfar, “Developing Press, 2014.
an integrated framework for using data mining techniques and ontology [28] E. Natalizio, Mobility as a Primitive to Improve Communications in
concepts for process improvement,” Journal of Systems and Software, Wireless Networks. PhD thesis, UTC Compiègne, 2017.
vol. 137, pp. 78–95, 2018. [29] D. Grigori, Eléments de flexibilité des systèmes de workflow pour la
[10] W. Van Der Aalst, A. Adriansyah, A. K. A. De Medeiros, F. Arcieri, définition et l’exécution de procédés coopératifs. PhD thesis, Université
T. Baier, T. Blickle, J. C. Bose, P. Van Den Brand, R. Brandtjen, Henri Poincaré-Nancy 1, 2001.
J. Buijs, et al., “Process mining manifesto,” in International Conference [30] H. Harrington, “P. american society for quality control quality, busi-
on Business Process Management, pp. 169–194, Springer, 2011. ness process improvement: the breakthrough strategy for total quality,
[11] S. Aguirre, C. Parra, and J. Alvarado, “Combination of process mining productivity, and competitiveness,” 1991.
and simulation techniques for business process redesign: a methodolog- [31] M. Dumas, M. La Rosa, J. Mendling, H. A. Reijers, et al., Fundamentals
ical approach,” in International Symposium on Data-Driven Process of business process management, vol. 1. Springer, 2013.
Discovery and Analysis, pp. 24–43, Springer, 2012.
[32] D. Grigori, F. Casati, M. Castellanos, U. Dayal, M. Sayal, and M.-C.
[12] D. Redlich and W. Gilani, “Event-driven process-centric performance Shan, “Business process intelligence,” Computers in industry, vol. 53,
prediction via simulation,” in International Conference on Business no. 3, pp. 321–343, 2004.
Process Management, pp. 473–478, Springer, 2011.
[33] B. Schwegmann, M. Matzner, and C. Janiesch, “A method and tool
[13] A. Yousfi, K. Batoulis, and M. Weske, “Achieving business process for predictive event-driven process analytics.,” in Wirtschaftsinformatik,
improvement via ubiquitous decision-aware business processes,” ACM p. 46, Citeseer, 2013.
Transactions on Internet Technology (TOIT), vol. 19, no. 1, pp. 1–19,
[34] W. Schmidt, “Business activity monitoring (bam),” in Business Intelli-
2019.
gence and Performance Management, pp. 229–242, Springer, 2013.
[14] C. Stoiber, “Exploiting internet of things for business process improve-
ment,” 2021. [35] O. Etzion and P. Niblett, “Event processing in action, manning publi-
cations,” 2011.
[15] M. M. Queiroz, S. F. Wamba, M. C. Machado, and R. Telles, “Smart
[36] R. von Ammon, O. Etzion, H. Ludwig, A. Paschke, and N. Stojanovic,
production systems drivers for business process management improve-
ment: An integrative framework,” Business Process Management Jour- “Introduction to the second international workshop on event-driven
nal, 2020. business process management (edbpm09),” in International Conference
on Business Process Management, pp. 345–346, Springer, 2009.
[16] S. Cherrier and V. Deshpande, “From bpm to iot,” in International
[37] C. Janiesch, M. Matzner, and O. Müller, “Beyond process monitoring: a
conference on business process management, pp. 310–318, Springer,
2017. proof-of-concept of event-driven business activity management,” Busi-
ness Process Management Journal, vol. 18, no. 4, pp. 625–643, 2012.
[17] R. Wehlitz, I. Rößner, and B. Franczyk, “Integrating smart devices
as business process resources–concept and software prototype,” in [38] D. Bo, D. Kun, and Z. Xiaoyi, “A high performance enterprise service
International Conference on Service-Oriented Computing, pp. 252–257, bus platform for complex event processing,” in 2008 Seventh Interna-
Springer, 2017. tional Conference on Grid and Cooperative Computing, pp. 577–582,
IEEE, 2008.
[18] S. Meyer, A. Ruppen, and C. Magerkurth, “Internet of things-aware pro-
[39] D. Luckham, “Soa, eda, bpm and cep are all complementary,” Retrieved
cess modeling: integrating iot devices as business process resources,” in
International conference on advanced information systems engineering, May, vol. 20, p. 2011, 2007.
pp. 84–98, Springer, 2013. [40] A. Estruch and J. A. H. Álvaro, “Event-driven manufacturing process
[19] C. Janiesch, A. Koschmider, M. Mecella, B. Weber, A. Burattin, management approach,” in International Conference on Business Pro-
cess Management, pp. 120–133, Springer, 2012.
C. Di Ciccio, G. Fortino, A. Gal, U. Kannengiesser, F. Leotta, et al.,
“The internet of things meets business process management: a mani- [41] R. v. Ammon, C. Emmersberger, F. Springer, and C. Wolff, “Event-
festo,” IEEE Systems, Man, and Cybernetics Magazine, vol. 6, no. 4, driven business process management and its practical application taking
pp. 34–44, 2020. the example of dhl,” 2008.
[20] S. Schönig, L. Ackermann, S. Jablonski, and A. Ermer, “An inte- [42] C. Costello, O. Molloy, and K. Hinkelmann, “Towards a semantic
grated architecture for iot-aware business process execution,” in Enter- framework for business activity monitoring and management.,” in AAAI
prise, business-process and information systems modeling, pp. 19–34, spring symposium: AI meets business rules and process management,
Springer, 2018. pp. 17–27, 2008.
[21] P. A. Stalenhoef, J. P. Diederiks, L. P. de Witte, K. H. Schiricke, and [43] D. Luckham, The power of events, vol. 204. Addison-Wesley Reading,
H. F. Crebolder, “Impact of gait problems and falls on functioning in 2002.
independent living persons of 55 years and over: a community survey,” [44] K. Danielsson and G. Trotta, “Key requirements for event-driven bpm
Patient Education and Counseling, vol. 36, no. 1, pp. 23–31, 1999. and soa.”
[22] P. A. Stalenhoef, J. P. Diederiks, J. A. Knottnerus, L. P. de Witte, [45] W. Yao, C.-H. Chu, and Z. Li, “Leveraging complex event processing
and H. F. Crebolder, “The construction of a patient record-based risk for smart hospitals using rfid,” Journal of Network and Computer
model for recurrent falls among elderly people living in the community,” Applications, vol. 34, no. 3, pp. 799–810, 2011.
Family Practice, vol. 17, no. 6, pp. 490–496, 2000. [46] R. Mousheimish, Y. Taher, K. Zeitouni, and M. Dubus, “Pact-art:
[23] P. Stalenhoef, J. Diederiks, J. Knottnerus, A. Kester, and H. Crebolder, Enrichment, data mining, and complex event processing in the internet
“A risk model for the prediction of recurrent falls in community- of cultural things,” in 2016 12th International Conference on Signal-
dwelling elderly: a prospective cohort study,” Journal of clinical epi- Image Technology & Internet-Based Systems (SITIS), pp. 476–483,
demiology, vol. 55, no. 11, pp. 1088–1094, 2002. IEEE.
[24] V. S. Stel, S. M. Pluijm, D. J. Deeg, J. H. Smit, L. M. Bouter, [47] Z. Maamar, E. Kajan, I. Guidara, L. Moctar-M’Baba, and M. Sellami,
and P. Lips, “A classification tree for predicting recurrent falling in “Bridging the gap between business processes and iot,” in Proceedings
community-dwelling older persons,” Journal of the American Geriatrics of the 24th Symposium on International Database Engineering &
Society, vol. 51, no. 10, pp. 1356–1364, 2003. Applications, pp. 1–10, 2020.
[25] K. A. Faulkner, M. S. Redfern, J. A. Cauley, D. P. Landsittel, S. A. [48] C. Friedow, M. Völker, and M. Hewelt, “Integrating iot devices into
Studenski, C. Rosano, E. M. Simonsick, T. B. Harris, R. I. Shorr, business processes,” in International conference on advanced informa-
H. N. Ayonayon, et al., “Multitasking: association between poorer tion systems engineering, pp. 265–277, Springer, 2018.
performance and a history of recurrent falls,” Journal of the American [49] R. Song, Y. Wang, W. Cui, J. Vanthienen, and L. Huang, “Towards
Geriatrics Society, vol. 55, no. 4, pp. 570–576, 2007. improving context interpretation in the iot paradigm: a solution to
[26] A. Furness, “A framework model for the internet of things,” integrate context information in process models,” in Proceedings of
GRIFS/CASAGRAS: Hong Kong, China, 2008. the 2018 2nd International Conference on Management Engineering,
Software Engineering and Service Sciences, pp. 223–228, 2018.
www.ijacsa.thesai.org 914 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

[50] S. Schönig, L. Ackermann, S. Jablonski, and A. Ermer, “Iot meets [69] D. Luckham, “A short history of complex event processing,” Part 1:
bpm: a bidirectional communication architecture for iot-aware process Beginnings. 1-a-short-history-of-cep-part-1. pdf, 2007.
execution,” Software and Systems Modeling, vol. 19, no. 6, pp. 1443– [70] D. Robins, “Complex event processing,” in Second International Work-
1459, 2020. shop on Education Technology and Computer Science. Wuhan, pp. 1–10,
[51] S. Schönig, L. Ackermann, and S. Jablonski, “Internet of things meets Citeseer, 2010.
bpm: A conceptual integration framework.,” in SIMULTECH, pp. 307– [71] G. Cugola and A. Margara, “Processing flows of information: From
314, 2018. data stream to complex event processing,” ACM Computing Surveys
[52] C. Sturm, S. Schönig, and C. Di Ciccio, “Distributed multi-perspective (CSUR), vol. 44, no. 3, p. 15, 2012.
declare discovery.,” in BPM (Demos), 2017. [72] D. S. Rosenblum and A. L. Wolf, A design framework for Internet-scale
[53] D. Evans, “The internet of things: How the next evolution of the internet event observation and notification, vol. 22. ACM, 1997.
is changing everything,” CISCO white paper, vol. 1, no. 2011, pp. 1–11, [73] M. Eckert and F. Bry, “Complex event processing (cep),”
2011. [74] A. Margara, G. Cugola, and G. Tamburrelli, “Learning from the past: au-
[54] Z. A. Almusaylim and N. Zaman, “A review on smart home present state tomated rule generation for complex event processing,” in Proceedings
and challenges: linked to context-awareness internet of things (iot),” of the 8th ACM International Conference on Distributed Event-Based
Wireless networks, vol. 25, no. 6, pp. 3193–3204, 2019. Systems, pp. 47–58, ACM, 2014.
[55] A. Ismaili-Alaoui, K. Baı̈na, K. Benali, and J. Baı̈na, “Towards smart [75] M. von Rosing and S. Polovina, “Business process trends,” 2015.
incident management under human resource constraints for an iot- [76] M. Daum, M. Götz, and J. Domaschka, “Integrating cep and bpm:
bpm hybrid architecture,” in International Conference on Web Services, how cep realizes functional requirements of bpm applications (industry
pp. 457–471, Springer, 2018. article),” in Proceedings of the 6th ACM International Conference on
[56] C. Soto, “Bpm and iot: The signal of true disruption,” Redmond Media Distributed Event-Based Systems, pp. 157–166, ACM, 2012.
Group, 2015. [77] M. Weidlich, H. Ziekow, J. Mendling, O. Günther, M. Weske, and
[57] C. Chang, S. N. Srirama, and R. Buyya, “Mobile cloud business N. Desai, “Event-based monitoring of process execution violations,”
process management system for the internet of things: a survey,” ACM in International conference on business process management, pp. 182–
Computing Surveys (CSUR), vol. 49, no. 4, p. 70, 2017. 198, Springer, 2011.
[58] J. Gubbi, R. Buyya, S. Marusic, and M. Palaniswami, “Internet of things [78] F. Koetter and M. Kochanowski, “A model-driven approach for
(iot): A vision, architectural elements, and future directions,” Future event-based business process monitoring,” Information Systems and e-
generation computer systems, vol. 29, no. 7, pp. 1645–1660, 2013. Business Management, vol. 13, no. 1, pp. 5–36, 2015.
[59] J. Pielmeier, S. Braunreuther, and G. Reinhart, “Approach for defining [79] P. Soffer, A. Hinze, A. Koschmider, H. Ziekow, C. Di Ciccio, B. Kold-
rules in the context of complex event processing,” Procedia CIRP, ehofe, O. Kopp, A. Jacobsen, J. Sürmeli, and W. Song, “From event
vol. 67, pp. 8–12, 2018. streams to process models and back: Challenges and opportunities,”
[60] D. C. Luckham, Event processing for business: organizing the real-time Information Systems, vol. 81, pp. 181–200, 2019.
enterprise. John Wiley & Sons, 2011. [80] B. Mazon-Olivo, D. Hernández-Rojas, J. Maza-Salinas, and A. Pan,
[61] D. Canty, “Architecting for dynamic analytics in iot..” https://dzone. “Rules engine and complex event processor in the context of internet
com/articles/architecting-for-dynamic-analytics-in-iot, 2014. [Online; of things for precision agriculture,” Computers and electronics in
accessed July-2021]. agriculture, vol. 154, pp. 347–360, 2018.
[62] S. Tjoa, S. Jakoubi, G. Goluch, G. Kitzler, S. Goluch, and G. Quirch- [81] N. Mehdiyev, J. Krumeich, D. Enke, D. Werth, and P. Loos, “Deter-
mayr, “A formal approach enabling risk-aware business process model- mination of rule patterns in complex event processing using machine
ing and simulation,” IEEE Transactions on Services Computing, vol. 4, learning techniques,” Procedia Computer Science, vol. 61, pp. 395–401,
no. 2, pp. 153–166, 2011. 2015.
[63] A. Ismaili-Alaoui, K. Baı̈na, and K. Benali, “Harnessing the power of [82] Q. Zhang, L. Cheng, and R. Boutaba, “Cloud computing: state-of-the-art
data and event data for business process improvement,” in CAISAM and research challenges,” Journal of internet services and applications,
2019-Complexity Analysis of Industrial Systems and Advanced Model- vol. 1, no. 1, pp. 7–18, 2010.
ing, 2019. [83] P. Mell, T. Grance, et al., “The nist definition of cloud computing,”
[64] U. ITIL, “Introducing service operation..” https://www.ucisa.ac.uk. [On- 2011.
line; accessed January-2021]. [84] J. Bell, Machine learning: hands-on for developers and technical
[65] J.Kaplan, “When internet of things meets big data.” professionals. John Wiley & Sons, 2020.
https://www.informationweek.com/big-data/big-data-analytics/ [85] J. A. Hartigan and M. A. Wong, “Algorithm as 136: A k-means
when-internet-of-things-meets-big-data/a/d-id/1298137?,. [Online; clustering algorithm,” Journal of the royal statistical society. series c
accessed July-2021]. (applied statistics), vol. 28, no. 1, pp. 100–108, 1979.
[66] J. Gutiérrez, V. Rodrı́guez, and S. Martin, “Comprehensive review of [86] R. L. Thorndike, “Who belongs in the family,” in Psychometrika,
vision-based fall detection systems,” Sensors, vol. 21, no. 3, p. 947, Citeseer, 1953.
2021. [87] P. J. Rousseeuw, “Silhouettes: a graphical aid to the interpretation and
[67] Y. Chen, X. Kong, L. Meng, and H. Tomiyama, “An edge computing validation of cluster analysis,” Journal of computational and applied
based fall detection system for elderly persons,” Procedia Computer mathematics, vol. 20, pp. 53–65, 1987.
Science, vol. 174, pp. 9–14, 2020. [88] R. Tibshirani, G. Walther, and T. Hastie, “Estimating the number of
[68] Cisco, “”cisco fog computing solutions: Unleash the power of clusters in a data set via the gap statistic,” Journal of the Royal
the internet of things”, syst. inc., san jose, ca, usa, white Statistical Society: Series B (Statistical Methodology), vol. 63, no. 2,
paper, 2015.” https://www.cisco.com/c/dam/en us/solutions/trends/iot/ pp. 411–423, 2001.
docs/computing-solutions.pdf. [Online; accessed July-2021].

www.ijacsa.thesai.org 915 | P a g e
 © 2022. This work is licensed under
https://creativecommons.org/licenses/by/4.0/ (the “License”). Notwithstanding
the ProQuest Terms and Conditions, you may use this content in accordance
with the terms of the License.