See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/363209825

The Interconnection of Internet of Things and Artificial Intelligence: A Review

Article · August 2022

CITATION READS

1 1,101

3 authors, including:

W.M.C.J.T. Kithulwatta R.M. Kapila Tharanga Rathnayaka

Uva Wellassa University Sabaragamuwa University of Sri Lanka
17 PUBLICATIONS 5 CITATIONS 82 PUBLICATIONS 438 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

novel forecasting mechanism on Time Series View project

VAR Model applications View project

All content following this page was uploaded by W.M.C.J.T. Kithulwatta on 02 September 2022.

The user has requested enhancement of the downloaded file.

 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

The Interconnection of Internet of Things and Artificial

Intelligence: A Review
M.H. Gayantha a,*, W.M.C.J.T. Kithulwattab,c, R.M.K.T. Rathnayakaa
a
Department of Physical Sciences and Technology, Faculty of Applied Sciences, Sabaragamuwa University of Sri Lanka,
Belihuloya, Sri Lanka, 70140
b
Faculty of Technological Studies, Uva Wellassa University of Sri Lanka, Badulla, Sri Lanka, 90000
c
Faculty of Graduate Studies, Sabaragamuwa University of Sri Lanka, Belihuloya, Sri Lanka, 70140

* Corresponding author email address: [email protected]

(Received 1st May 2022; accepted 08th August 2021)

Abstract

The current world is moving forward with digitalization. In this era, new trends such as artificial intelligence (AI), machine
learning (ML), the internet of things (IoT), and data analytics technologies became popular among people. People are trying to
make their work easier using machines or other ways. The autonomous concept is another idea that became based on the above-
mentioned trends. In the health service, agricultural sector, and business sector this IoT and AI were widely used. Therefore,
AI and IoT interconnected applications become a good trend these days. Furthermore, lots of industries change their manual or
electrical logistics into IoT-based platforms that can be more easily manageable, and more accurate than before. The reliability
of that systems is also very high because AI is given the power to that systems. Therefore humans do not need to pay attention
to lots of tasks because AI-powered IoT platforms take care of the system. This paper reviewed previous literature, and scholarly
articles to identify what is the interconnection of AI and IoT. Twenty-three existing research studies were used for this study.
This paper describes AI and IoT, the interconnectivity between both skills, the challenges of AI and IoT applications, and what
are the fields those applications used and uses, from other researchers’ perspective. This research study brings a collection of
domain knowledge to the computing discipline regarding the integration of IoT and AI.

Keywords: Artificial Intelligence (AI), Data Analytics, Internet of Things (IoT), Machine Learning (ML)

1. Introduction member of the RFID (Radio Frequency Identification)

community [4]. The concept is based on the connection of
Artificial intelligence (AI) and the internet of things (IoT) physical devices over the internet to do a specific task or
become trending topics in the world among other multiple tasks, and monitoring sensor values to get real-time
technologies with the enhancement of technology in the data to do the above specific tasks. The physical devices are
digitalized world. Humans try to find out new things quickly connected to a microcontroller board and it is the heart of the
and try to reduce the effort that they put into their regular IoT system. The system can control or monitor from
work and try to convert the human brain into an artificial one anywhere in the world. System monitoring and analyzing
that can work the same as a human brain. In the modern era, devices can be desktop computers, laptop computers, mobile
AI is being become a conspicuous trend in the industries phones, tablets, or specific screens. IoT can be specified as a
such as construction, software development, health, global setup that allows the communication between human-
marketing, logistics, transportation, education, agriculture, to-human, human-to-things, and things-to-things.
etc [1]. AI is a powerful tool that can use to make decisions Communication between devices can happen through
using simulations and modules [2]. AI-powered devices and wireless or wired media. IoT is being gone to be everywhere
tools are more smart and able to do a particular task that around the human environment. IoT applications can be used
saves time and resources [3]. AI can be divided into several in different fields such as agriculture, medicine,
parts which are learning, reasoning, problem-solving, transportation, manufacturing, education, and people's day-
perception, and natural language understanding, and to-day operations. Home automation systems, smart cities,
researchers have used AI for computer vision, speech IoT retail shops, auto-driven cars, and farming are some
recognition, natural language processing, expert systems, applications that are used IoT.
game playing, intelligent robots, etc. As well, machine This research paper consists of a comprehensive review
learning (ML), neural networks, and deep learning are the of the interconnection of AI and the IoT. The paper focused
core of AI concepts. to identify what are the applications developed using IoT
IoT has also become a hot trend in the last ten years. The combined with AI, what are the fields that used IoT and AI,
IoT concept was born in 1999, and it was introduced by a and the challenges faced in IoT and AI.

50
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

This research study brings answers to the below research title, abstract,
questions(RQs). and conclusions
were read each
RQ1: What is the compatibility of AI and IoT? paper. From this
RQ2: What are the economic advantages of AI and IoT selection, 23
integration? papers were
RQ3: What are the challenges of AI and IoT integrations? finalized for this
review.
07 Synthesizing On top of the
2. Research Methodology selected papers,
synthesizing was
To conduct this literature review study, it was followed a applied.
highly structured review technique. The technique includes 08 Reporting By reading and
eight steps. Table 1 describes the steps that were selected to exploring the
conduct this review. Table 1 has three columns: the first selected
column is the step number, the second column is the step literature, final
name, and the third column is described the step function. observations and
results were
Table 1 identified and
Research Methodology Steps reported.
Step Step Name Step
Number Description The search string:
01 Reason for review Identify the {[(AI*) ˄ (IoT*)] ˅ Interconnection*}; for *,
reason for the lexicographically suitable words were used.
review and it is
described in the The selected search string was queried in Google Scholar.
research Google Scholar was used as the key search engine to find
identification. literature. Mainly, research studies were selected from IEEE
02 Research Questions To conduct the Digital Explore, Springer, Science Direct, Hindawi, Willey
review in a Online Library, Uppsala University Repository, and other
structured way, databases.
the research After the selection of initial papers: the selected papers
questions were were screened by viewing the title of the paper, keywords of
identified. the paper, and abstract of the paper. The studies which are
outside of the domain of AI and IoT were excluded from the
03 Select the search string The search string
initial collection.
is used to select
literature. To conduct this review study, the following research
studies were selected. Below table 2 describes the titles of
04 Initial literature By using the
the selected literature and column one describes the citation
selection. search string,
number and column two describe the topic of the selected
literature was
research paper.
selected using
Google Scholar
Table 2
from different
databases such as Titles of Referred Articles
IEEE-Xplore, [1] IoT‐Enabled AI‐Based Model to Assess Land
Springer, Suitability for Crop Production
Science Direct, [2] A Survey of Artificial Intelligence and
and many Internet of Things (IoT) based approaches
journals. against Covid-19
05 Inclusion/ Exclusion To filter the [3] AI-Enabled Sensing and Decision-Making for
papers from the IoT Systems
domain, the [4] Synergy of IoT and AI in Modern Society:
paper inclusion The Robotics and Automation Case
and exclusion [5] A Review of Artificial Intelligence in the
criteria were Internet of Things
applied [6] Artificial Intelligence in Internet of Things
06 Quality Assessment To select correct [7] Challenges in the Integration of Artificial
and suitable Intelligence and Internet of Things
papers, the paper

51
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

[8] Internet of Things (IoT): A Literature Review Minitab statistical software. According to that, most of the
[9] Role of Artificial Intelligence in the Internet papers were published as journal articles.
of Things – Review Table 5 presents a statistical summary of the selected
[10] Review on Artificial Intelligence with Internet literature for publication year. The most of selected research
of Things- Problems, Challenges and studies were published in the year 2020. For the research
opportunities study, the research articles were selected within the last
[11] Research on Artificial Intelligence Enhancing seven years.
Internet of Things Security: A Survey
[12] The Relation Of Artificial Intelligence With Pie Chart for Publication Format Distribution
Internet Of Things: A survey Category
Book Chapters
[13] Application of AI and IoT in Clinical 8.7% 4.3% Conference Papers
Journal Articles
Medicine: Summary and Challenges 4.3%
17.4%
Thesis
Survey

[14] Survey on IoT security: Challenges and

solution using machine learning, artificial
intelligence and blockchain technology
[15] A comprehensive review on automation in
agriculture using artificial intelligence
[16] Technology assisted farming: Implications of
IoT and AI
[17] Survey on Machine Learning and Deep 65.2%

Learning Algorithms used in Internet of

Things (IoT) Healthcare
[18] A Survey on the Internet of Things (IoT) Fig. 1. Pie Chart for Distribution of Publication Format
Forensics: Challenges, Approaches and Open
Issues Table 4
[19] Business models based on IoT, AI and Research Publication Distribution according to the Year
blockchain Publication Year Number of Papers
[20] Transforming Business Decision Making with The year 2021 5
Internet of Things (IoT) and Machine The year 2020 8
Learning (ML)
The year 2019 4
[21] Internet of Things (IoT)-Enabled Unmanned
The year 2018 4
Aerial Vehicles for the Inspection of
The year 2016 1
Construction Sites: A Vision and Future
The year 2015 1
Directions
[22] Machine Learning in IoT Security: Current
3. Artificial Intelligence
Solutions and Future Challenges
[23] A survey of machine learning for big data
In the review on “Artificial Intelligence in the Internet of
processing
things”, the authors have highlighted what is AI and a few
main branches of AI with their concepts, process, and related
Below table 3 presents a statistical summary of the selected
work on IoT [6]. According to their paper, AI was created to
literature for publication format. The selected research
allow programmers to create programs that allow computers
studies were published as book chapters, journal articles,
to learn, and AI is the science and engineering that attempts
conference papers, thesis, and survey articles. Table 3 is with
to make machines intelligent by teaching them to understand
two columns: the first column denotes the publication format
human language and solve problems and achieve goals in the
and the second column denotes the number of papers for
same way that humans do. Automatic learning, computer
each category.
vision, fuzzy logic, natural language processing, heuristics,
and intelligent agents are some of the fields of AI that may
Table 3
be split. They defined six rules in AI;
Research Publication Format
Publication Format Number of Papers 1. It must design not to exceed humanity
Book Chapters 01
2. It must be transparent about the technology and
Conference Papers 4
Journal Articles 15 rules.
Thesis 1 3. It should be maximizing the effectiveness
Survey 2
4. It should have algorithmic responsibility
To visualize the selected paper distribution among the set 5. It designed for intelligent privacy
of papers, the below pie chart was illustrated by using

52
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

6. It must not be biased [5] with other devices and structures through the internet. From
simple nuclear family articles to complicated modern
The researchers have answered the issue with AI and IoT, devices, these tactics are used. IoT allows devices/items to
which rises when handling huge data generated with the least monitor, identify, and understand a situation or
power computers. According to the paper [2], AI is the study environmental conditions without the need for human
of imbuing intelligence into computers so that they may do intervention. From everyday items like kitchen equipment,
activities that previously required the use of the human autos, indoor regulators, and children's screens to the
intellect. In terms of applicability, adaption, processing internet through implanted devices. IoT is widely employed
speed, and capabilities, AI-based systems are quickly in many officialdoms, just as AI expertise [9].
evolving. Machines are growing capable of performing jobs The phrase “IoT” is already widely used, yet there is no
that aren't as regular. While human intelligence is defined as universally accepted definition or understanding of what it
"taking" a flawless decision at the right moment, AI is means. Kevin Ashton, executive of AutoIDCenter (MIT),
defined as "choosing" a correct decision at the right time. To coined the phrase "IoT" in 1999 while chipping away at the
put it bluntly, AI cannot make decisions with the same roots of structured "radio recurrence distinguishing proof"
ingenuity as humans. It may be claimed that human (RFID). He coined the phrase to reflect his vision of a society
inventiveness will always modify the function of productive in which every electronic device is organized, and
labor, but AI-based systems have ingeniously decreased everything (physical or electronic) is tagged with
human effort repeatedly and can provide outcomes in a very information specific to that item. IoT also known as the
short amount of time. The majority of current AI research Internet of Objects (IoO), is a development of the internet in
may be classified as 'Narrow AI.' This implies that which the things/objects distinguish themselves by
technology can only help with jobs. As a result, numerous transmitting data about them. They can acquire data about
areas have converged to propel AI forward [6]. themselves from various products and things, or they can be
According to the research paper [7], AI is a technology part of higher-level administrations. [10]
that enables us to make any computer intelligent, a IoT is evolving and remains the most anticipated
computer-driven robot, or software that thinks intelligently information technology concept. According to the research
in the same manner that a person does. AI is becoming a paper [8], the phrase “IoT” has gained popularity over the
more advanced technology day by day. AI is a human last decade by portraying the idea of a worldwide
intelligence process that will be felt by machines. ML and infrastructure of networked physical things that enables
deep learning are AI programs that can automatically learn anytime, everywhere connection for anything and not just
from experience and enhance performance. Deep learning is everyone. IoT is a worldwide network that permits
a subset of ML that is made up of algorithms that teach it communication between human-to-human, human-to-
how to do tasks such as speech recognition and face things, and things-to-things, which is everything in the
identification with enormous volumes of data [7]. world, by giving each object a unique identity. IoT portrays
According to the [8] research paper, AI refers to is a world in which almost everything may be connected and
electronic settings that are aware of and respond to human interacts in a more sophisticated manner than ever before.
presence. In an ambient intelligence environment, Most of us associate "connection" with technological
technologies work together to assist individuals in carrying equipment such as servers, laptops, tablets, phones, and
out their daily tasks in a simple, natural manner by utilizing smartphones. Sensors and actuators embedded in real objects
information and intelligence concealed in networked devices from highways to pacemakers are linked over wired and
[8]. wireless networks in what's known as the IoT, which
The discipline of AI is concerned with intelligent typically uses the same internet protocol (IP) that links the
machines, or rather, with embedding intelligence into internet [8].
computers, since "artificial intelligence is the science and Applications of IoT were listed below according to the
engineering of producing intelligent machines." AI has research paper [7].
evolved into a critical component of the computer industry,
assisting in the resolution of some of society's most complex 1. Smart houses
issues. Expert systems, or computer programs that imitate 2. Wearable’s
the logic and performance of human experts, are one type of
AI. On the other hand, may be defined as a computer 3. Connected cars
program that solves complicated issues that would normally 4. Industrial internet
need a great deal of human knowledge. To do this, it uses
rules or objects to reflect human knowledge to imitate the 5. Smart cities
human reasoning issue [4]. 6. IoT in Agriculture
4. IoT 7. Smart retails
8. IoT in health care
IoT is a collection of physical objects “things” that are
integrated with sensors, programming, and numerous 9. Energy management
interfaces to communicate with and replace the information

53
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

The concept "Internet of Computers (IoC)" has existed decision algorithms, transaction algorithms employ limited
since 1991 and has progressively grown in size as more samples and basic models. Decision algorithms are primarily
people began to use it. The Internet of Gadgets (IoG) began responsible for business choices and employ various
with the introduction of pocket phones and linked devices, decision-making processes to lower the ratio of misjudgment
and as mobile phones, computers, laptops, and tablets and maximize total profit. Single decision-making,
became more affordable and accessible to the general public, sequential decision-making, and integrated decision-making
the internet of devices increased in size. In 2016, there were are three types of decision algorithms that may be classified
around 6.4 billion linked items on the planet, up 30% over by methods and situations [11].
2015, and 20.8 billion by 2020. In 2016, almost 5.5 million According to the research paper [9], comprehending the
new items were linked every day, demonstrating the IoT’s role of AI in the IoT revolution was discussed. Simulated
enormous potential. Because numerous devices are intelligence is expected to do many intelligent tasks, such as
constantly linking to build an IoT, it is related to a variety of speech recognition, language interpretation, dynamics, and
specialties. As a result, the IoT may be conceived of as a so on, without the need for human intervention. The IoT, on
collection of different domains. The IoT is a network of the other hand, connects a network of networked devices that
physical objects and humans. Humans are linked to these exchange data across a network. IoT gadgets have found
gadgets via smart items that can send, receive, and analyze their way into our daily lives and are hoped to provide a
data and are tied to both. These smart objects reflect the higher degree of comfort. These contraptions rely on web
entity in the network to which they are connected. [9] structure and generate unimaginable amounts of data that
may be applied to client behaviors, tendencies, unique
5. The connection between AI and IoT information, and so on, and so cannot be ignored.
Nonetheless, many endeavors are unaware of how to store
In a lot of research articles, AI-enabled IoT was discussed and cycle such massive amounts of data. The growth and
in many main topics. IoT is a huge idea that encompasses so capacity of IoT are being hampered because of this. For this
many detectors, actuators, data storage, and data processing situation, the man-made intellectual capability may be a
capabilities, all of which are interconnected through the huge assistance in sifting through the deluge of data
internet. As a result, every IoT-enabled gadget may detect its generated by IoT devices. It allows you to examine facts and
surroundings, transmit, store, and interpret the data it make good decisions based on what you find [9].
collects, and take appropriate action. The last phase of acting The relationship of AI with IoT was discussed with some
in accordance is completely reliant on the processing stage. applications as below.
The degree of processing or action that an IoT service can
do determines its genuine smartness. A non-smart IoT 1. Home Automation
system will have restricted capabilities and will not be able
2. Oil Field Production
to adapt in tandem with the data. A better IoT system, on the
other hand, will contain AI and may be able to achieve the 3. Smart Hotel
real aim of automation and adaptation [6]. The below-
mentioned examples were elaborated in their paper which AI algorithms and approaches can evaluate and learn
combined context with IoT and AI. from the massive amounts of data generated by linked IoT
devices, allowing public services and value to be created.
1. Voice assistants such as Alexa, Siri, and Google AI's rising popularity, which gives superior data analytics
Assistant choices, has a significant impact on IoT usage. The
discipline of AI is concerned with intelligent machines, or
2. Robots like Pepper, Sophia rather, with embedding intelligence into computers, i.e.,
3. Smart devices; Smart Oven, SkyBell "artificial intelligence is the science and engineering of
producing intelligent machines". AI is the study of machine
4. Industrial IoT like primer, Plutoshift intelligence as well as the discipline of computer science that
seeks to develop it [12].
The research paper [11] has mentioned the security of IoT
and how it can be enhanced with AI. Because of the unique 6. The connection between AI and IoT and challenges
nature of IoT security and the limits of existing solutions, in IoT and AI
new security technologies are urgently required. AI offers a
broad variety of applications as a new technology path. In IoT and AI each have their own set of issues; when these
the realm of AI, ML is a study topic. Its theory and methods two technologies are combined, the challenges become
have been used to a wide range of engineering challenges to much more complicated. Some of the issues include the
address complicated problems. Transaction and decision following [7].
algorithms are two types of ML algorithms that are used in
IoT security [11]. 1. Security
The primary functions of transaction algorithms are
2. Compatibility and complexity
information discovery and data preparation. To collect the
general features of the dataset and give the foundation for 3. Artificial Stupidity

54
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

4. Lack of Confidence system for flower and leaf identification and watering that
can be implemented in a botanical farm using IoT [11].
5. Cloud Attacks
The authors of the research paper [17] discuss the
6. Technology technology-assisted farming implemented by IoT and AI.
The purpose of this article is to explain how the IoT has
In the research paper [13], the authors have discussed the changed the agriculture industry. According to the survey,
applications of IoT and AI in clinical medicine with a 70% of India's population relies on agriculture for a living,
summary and challenges. Intelligent medical IoT exhibits a but agriculture's position is no longer hidden from society.
variety of new capabilities as a result of the combination of Temperature, rainfall, humidity, fertilizer requirements,
AI, IoT, and the medical business. IoT connection terminals water requirements, and other factors may all be predicted
provide a wide range of data modalities, which come in a with the use of technology. Modern agricultural approaches
variety of formats and have cross-domain fusion capabilities. based on IoT, and AI are altering old agriculture practices
As a result, multidisciplinary knowledge fusion is required while also making farming lucrative [18].
in this sector, such as medical, AI, big data, information
engineering, and so on. New features, on the other hand, 7.2 Health sector
provide new obstacles, and smart medical IoT faces several
dangers and challenges [13]. Covid-19 is a trending factor in current years from 2020.
The authors of the research paper [2] have conducted a
1. Information Security is Threatened survey of AI and IoT based on approaches against Covid-19
which is related to the health sector. This study examines a
2. IoT Data Standards are Missing
variety of AI applications and strategies that can help us
3. Risks Brought by the Refinement of Specialties defeat Covid-19 and win the war. Computer vision, speech
recognition, natural language processing, and data analytics
4. Impact of AI on Ethics
are just a few examples of AI application subdomains. The
5. Medical Compound Talents are Lacking report identifies areas where AI and the IoT combine to give
effective methods for preventing, diagnosing, and reducing
In the research article [18], the authors began by Covid19. The report also includes a review of research on
providing an overview of IoT technology and its application technological applications against Covid-19 and data
areas. Security issues were discussed using ML, AI, and sources connected to it, as well as a survey of research efforts
blockchain technology. In the conclusion, they have and methods on these topics. The authors also suggest a
mentioned an analysis of issues. In recent years, the IoT has unique method for detecting Covid-19 in its early stages,
got a lot of interest from both the scientific community and which combines AI-enabled IoT sensors with conversational
the business community. The IoT gadgets are being capabilities.
produced in massive quantities, and they have already Another Survey on ML and deep learning used in IoT
reached the global population. These smart gadgets are healthcare. This research attempts to comprehend in depth
linked to a variety of apps that collect data from their the algorithms that have the potential to enhance healthcare
surroundings. Because IoT devices have limited resources, systems based on IoT, the majority of which have already
they are susceptible to assault. IoT applications, security, been deployed. Because of the reliance on AI and deep
and privacy are crucial [18]. learning, there is a significant likelihood of a human mistake.
The performance of training algorithms is improved when
7. IoT Applications Powered by AI standardized data sets are used instead of unstandardized
data sets, which reduces complexity and computation time
7.1 Agricultural sector [16].
The healthcare industry is perhaps the most vulnerable to
The research article [14] discusses several automation serious security breaches of all IoT-based domains. This
techniques such as IoT, wireless communications, ML, AI, might be explained by the fact that IoT applications in this
and deep learning. Crop diseases, lack of storage industry are cross-organizational, as well as their
management, pesticide control, weed management, lack of heterogeneity, fragmentation, and enlarged attack surface.
irrigation, and water management are some of the issues that As a result, although remote health monitoring technologies
plague the agriculture industry, and all of these issues may are revolutionizing the healthcare business and enhancing
be remedied utilizing the approaches stated above. Today, it people's lives, they also raise new questions concerning the
is critical to decipher concerns such as the use of dangerous privacy and security of users' medical data. Malicious actors,
pesticides, regulated irrigation, pollution management, and for example, might attack fitness trackers to profit from the
environmental repercussions in agricultural activity. information acquired. Wearables, such as fitness wristbands,
Automation of farming processes has been shown to boost have also become popular in forensics as a source of digital
soil gain while also strengthening soil fertility. This paper evidence [2].
compiles the findings of several researchers to provide a
quick overview of how automation is being implemented in
agriculture today. This paper also discusses a proposed

55
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

7.3 Business

According to the master thesis [19], the major goal is to

show how to think about integrating these new technologies
into business models and evaluating the most important parts
during the business planning stage. It has the potential to
assist future companies in seizing opportunities and
overcoming problems in the new business climate. To enable
physical asset transactions on the blockchain, a solution to
connect online and offline resources must be developed,
which is the essential notion of a blockchain-based supply
chain system. Even though several theories have been
proposed, the definitive solution has yet to be found.
Fig. 2. Three Constraints of IoT [21]
Because blockchain will become the backbone for
practically all sorts of transactions in the future,
understanding such patterns might help entrepreneurs better With automated intelligence, the three conditions listed
develop their business models. In addition, the sample pool below must be met to provide accurate results: The scalable
is quite tiny due to time and resource constraints. In the network architecture sometimes referred to as network
future, a larger sample pool will be preferable to produce a scalability, should be established. Intelligence is defined as
more precise and thorough study [19]. ensuring the best computing algorithm that quickly extracts
The author of the paper [20] is devoted to enumerating the data and determines the outcome. support for a variety of
and presenting the most recent problems in implementing sensors to collect data is known as diversity. The above
disruptive technology in business operations. The authors figure 2 presents the diagram for the main IoT constraints for
have shown an architectural model for an IoT and ML-based reliable vehicle-based systems [21].
application that can be linked with a variety of different
corporate apps to provide real-time data analytics, visibility, 8 Machine learning techniques for IoT security in IoT
and decision-making. A few IoT and ML-based commercial networks
applications in areas such as manufacturing,
pharmaceuticals, and hospitality have been briefly covered. ML algorithms can be divided into four main categories.
Finally, the authors have outlined the primary issues and These are supervised, unsupervised, semi-supervised, and
potential remedies [20]. reinforcement learning algorithms. Under each category
many example algorithms are available. The IoT network's
core security and privacy challenges are the focus of ML
7.4 Unmanned aerial vehicle
techniques. More specifically: malware analysis, distributed
denial of service (DDoS) assaults, anomaly and intrusion
IoT is described as the intelligent linking of things to the detection, and attack detection and mitigation were taken
internet using some active sensors. Better building site
into the account [22] [23].
inspection method using an IoT-enabled drone. By doing so,
In IoT networks, supervised learning techniques are used
one can also handle the limitations of the IoT, such as
to solve challenges related to spectrum sensing, channel
network scalability, processing limitations, and various
estimation, adaptive filtering, security, and location by using
provisions for various applications. Unmanned aerial labeled data. Regression and classification are two different
vehicles with IoT capabilities could be proposed as a way to sorts of approaches that fall under this category. For
not only overcome these limitations but also improve these
prediction and modeling of the available data sets,
three crucial IoT expectations. IoT has exploded in recent
classification under supervised ML is performed. To predict
years and is now heavily utilized across several industries.
continuous numeric variables, regression is employed. Some
The IoT idea, put simply, refers to connecting systems with
of the popular classification algorithms are Support vector
various internet capabilities. Therefore, it chooses machine- machines (SVM), Naive Bayes, Random Forest, and
to-machine communication the majority of the time.
Decision Trees. To determine the difference between two
Therefore, the research suggests connecting the drone to a
points belonging to two different classes, SVM employs a
base station system where a construction site supervisor is
method known as kernels. The non-linear decision
monitoring the status of work [21].
boundaries can be modeled by SVMs. It becomes
challenging to model huge datasets with SVM since it is
inherently memory-intensive, difficult to choose an
appropriate kernel, and memory-intensive. Random forests
are typically favored over SVM as a result. While Naive
Bayes (NB) is employed to simulate actual issues like spam
detection and text classification. Random forest algorithms
are perfect for simulating real-world issues since they are
naive and assume that all input features are independent of
one another [22].

56
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

Random forest methods can adjust to the size of the given As a result, deep learning-based algorithms outperform
data set and are simpler to implement. Compared to other traditional theories and methods while having lower latency
supervised algorithms like SVM and NB, these methods and complexity. Deep neural networks are also proficient at
require more time to train. However, it is more accurate and identifying and defining low-dimensional representations
requires less time to anticipate. Additionally, it is built on from any kind of high-dimensional data patterns (text,
building a graph with leaves and branches that represent image, or audio). In heterogeneous IoT networks, DDoS
decision and class, respectively. A top-down strategy is used detection and authentication are accomplished via deep
to categorize an event by iterating through the tree until a reinforcement learning and its derivatives. Deep
class is selected. The two most well-known regression deterministic policy gradient, continuous deep Q-Network,
techniques are logistic regression and nearest neighbors. prioritized experience replay, asynchronous N-step Q-
These algorithms, also referred to as "instance-based," use learning, deep SARSA, Dueling network deep Q-Network,
the most comparable training data to generate predictions for and asynchronous advantage actor-critic are the main deep
each new observation. But these algorithms struggle with reinforcement learning algorithms used for security and
high-dimensional data and require a lot of memory [22]. privacy [22].
The group of unsupervised learning algorithms works Table 5 presents an overview of security challenges in
with unlabeled data and makes heuristic use of input data. IoT networks and applied ML and deep learning techniques
These are employed in load balancing, cell clustering, to detect the particular situation according to the [22].
anomaly, fault, and intrusion detection. Data groupings
based on some inherent similarities and differences are done Table 5
using clustering, which falls under the topic of unsupervised IoT network security issues and applied machine learning
learning. There are no right or wrong answers because the techniques
clustering is unsupervised. Data visualization is employed to Objective Techniques for Machine
assess the correctness of the results. The clusters in datasets
Learning
can be pre-labeled if there is a possibility for a correct or
Authentication ▪ Deep Learning
incorrect answer, and in this case, classification methods are
▪ Recurrent neural
preferable. Popular clustering algorithms include
networks
hierarchical clustering and K-means. The most often used
▪ Q-learning and Dyna-Q
algorithm is K-Means, which creates clusters based on the
▪ Deep Neural Network
geometric separations between data points. Around
Attack Detection ▪ SVM
centroids, clusters assemble to form globular structures of
and Mitigation ▪ Deep Learning
the same size. Before clustering, the number of clusters must
▪ Unsupervised learning,
be determined, however, this is not always feasible or
stacked autoencoders
efficient. Additionally, cluster formation is weak if clusters
▪ Extreme Learning
are not spherical. The majority of IoT applications use
Machine (ELM)-based
unsupervised learning techniques with very little baseline
semisupervised Fuzzy C-
environmental knowledge, which is comparable to how to
Means (ESFCM)
live creatures naturally learn. For instance, zero-day attacks
▪ K-Nearest Neighbour
on IoT networks typically start with little to no information
and SVM
[22].
Distributed DOS ▪ K-Nearest Neighbour
Reinforce learning methods develop the reward and
action link between the agent and the environment by taking Attack ▪ Support Vector Machine
advantage of different stages. This action-reward ▪ Random Forest and
Decision Tree
relationship is highly helpful in addressing a variety of IoT
▪ Neural Network
issues. It does not require a large training data set, but the
▪ Multivariate Correlation
agent must be familiar with the state transition function.
Despite being computationally straightforward, reaching a Analysis (MCA)
steady state takes a long time. Utilizing reinforce learning ▪ Q learning
techniques in the dynamic settings of IoT networks presents Anomaly/Intrusion ▪ K-means clustering and
several difficulties, including the sluggish convergence and Detection Decision Tree
understanding of the state transition function or optimal ▪ Artificial Neural Network
policy [22]. ▪ Novelty and Outlier
Security and privacy are just two of the key areas in the Detection
IoT domain where deep learning might offer more effective ▪ Decision Tree
solutions. Deep learning is dependent on strong function ▪ Naive Bayes
approximation, estimate, and learning capabilities. Due to Malware Analysis ▪ Recurrent Neural
their resource limitations, IoT devices might not be able to Network (RNN)
host or perform sophisticated computational algorithms for ▪ ensemble learning
tasks like communication, analysis, and prediction [22]. algorithm Random Forest
supervised classifier

57
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

▪ Deep Eigensapce Available: 10.19080/raej.2018.03.555621 [Accessed 6

Learning and Deep August 2022].
Convolutional Networks
▪ SVM [5] C. González García, E. Núñez-Valdez, V. García-Díaz,
▪ PCA (Principal C. Pelayo G-Bustelo and J. Cueva-Lovelle, "A Review of
Component Analysis), Artificial Intelligence in the Internet of Things",
one-class SVM, and International Journal of Interactive Multimedia and
naive anomaly detector Artificial Intelligence, vol. 5, no. 4, p. 9, 2019. Available:
based on unseen n-grams 10.9781/ijimai.2018.03.004 [Accessed 6 August 2022].
▪ CNN (Convolutional
Neural Network) [6] A. Ghosh, D. Chakraborty and A. Law, "Artificial
▪ Artificial Neural Network intelligence in Internet of things", CAAI Transactions on
▪ Linear SVM Intelligence Technology, vol. 3, no. 4, pp. 208-218, 2018.
▪ SVM and PCA Available: 10.1049/trit.2018.1008 [Accessed 6 August
2022].
9 Conclusion
[7] G. Katare, G. Padihar and Z. Qureshi, "Challenges in the
IoT and AI are both trending topics in recent years. IoT Integration of Artificial Intelligence and Internet of Things",
and AI are both extremely strong technologies on their own. International Journal of System and Software Engineering,
IoT – AI is created when AI and IoT are combined. AI is the vol. 6, no. 2, 2022. [Accessed 6 August 2022].
brain of a system, while IoT devices are the digital nervous
system. In this review paper, AI and IoT were defined from [8] S. Madakam, R. Ramaswamy and S. Tripathi, "Internet
a few authors’ papers. The main part of this paper was the of Things (IoT): A Literature Review", Journal of Computer
interconnection between IoT and AI. and Communications, vol. 03, no. 05, pp. 164-173, 2015.
The research study has brought the answers for RQ1, Available: 10.4236/jcc.2015.35021 [Accessed 6 August
RQ2, and RQ3 respectively within the discussion. The 2022].
research study has identified that the integration of AI and
IoT is more powerful for day-to-day operations of real life [9] M. Kuzlu, C. Fair and O. Guler, "Role of Artificial
and organizational functionalities. By considering the Intelligence in the Internet of Things (IoT) cybersecurity",
business approaches, agriculture, education, health, vehicle Discover Internet of Things, vol. 1, no. 1, 2021. Available:
autonomy, fuel production, and smart cities are the most 10.1007/s43926-020-00001-4 [Accessed 6 August 2022].
economical points for enhancing the capacities with IoT and
AI integration. [10] A. K. Rana, R. Krishna, S. Dhwan, S. Sharma and R.
Gupta, "Review on Artificial Intelligence with Internet of
References Things - Problems, Challenges and Opportunities," 2019 2nd
International Conference on Power Energy, Environment
[1] A. Gudavalli, G. JayaLakshmi and S. Manne, "IoT‐ and Intelligent Control (PEEIC), 2019, pp. 383-387, doi:
Enabled AI‐Based Model to Assess Land Suitability for 10.1109/PEEIC47157.2019.8976588.
Crop Production", Green Internet of Things and Machine
Learning, pp. 215-237, 2021. Available: [11] H. Wu, H. Han, X. Wang and S. Sun, "Research on
10.1002/9781119793144.ch8 [Accessed 6 August 2022]. Artificial Intelligence Enhancing Internet of Things
Security: A Survey," in IEEE Access, vol. 8, pp. 153826-
[2] O. Nadeem, M. S. Saeed, M. A. Tahir and R. Mumtaz, 153848, 2020, doi: 10.1109/ACCESS.2020.3018170.
"A Survey of Artificial Intelligence and Internet of Things
(IoT) based approaches against Covid-19," 2020 IEEE 17th [12] E. Mohamed, "The Relation Of Artificial Intelligence
International Conference on Smart Communities: Improving With Internet Of Things: A survey", ournal of Cybersecurity
Quality of Life Using ICT, IoT and AI (HONET), 2020, pp. and Information Management (JCIM), vol. 1, no. 1, 2020.
214-218, doi: 10.1109/HONET50430.2020.9322829. Available: 10.5281/zenodo.3686810 [Accessed 6 August
2022].
[3] H. Qinxia, S. Nazir, M. Li, H. Ullah Khan, W. Lianlian
and S. Ahmad, "AI-Enabled Sensing and Decision-Making [13] Z. Lu et al., "Application of AI and IoT in Clinical
for IoT Systems", Complexity, vol. 2021, pp. 1-9, 2021. Medicine: Summary and Challenges", Current Medical
Available: 10.1155/2021/6616279 [Accessed 6 August Science, vol. 41, no. 6, pp. 1134-1150, 2021. Available:
2022]. 10.1007/s11596-021-2486-z [Accessed 6 August 2022].

[4] S. G Tzafestas, "Synergy of IoT and AI in Modern [14] B. Mohanta, D. Jena, U. Satapathy and S. Patnaik,
Society: The Robotics and Automation Case", Robotics "Survey on IoT security: Challenges and solution using
& Automation Engineering Journal, vol. 3, no. 5, 2018. machine learning, artificial intelligence and blockchain
technology", Internet of Things, vol. 11, p. 100227, 2020.

58
 Sri Lankan Journal of Applied Sciences Vol.1.1 (2022) 50-59

Available: 10.1016/j.iot.2020.100227 [Accessed 6 August

2022].

[15] K. Jha, A. Doshi, P. Patel and M. Shah, "A

comprehensive review on automation in agriculture using
artificial intelligence", Artificial Intelligence in Agriculture,
vol. 2, pp. 1-12, 2019. Available: 10.1016/j.aiia.2019.05.004
[Accessed 6 August 2022].

[16] N. Aggarwal and D. Singh, "Technology assisted

farming: Implications of IoT and AI", IOP Conference
Series: Materials Science and Engineering, vol. 1022, no. 1,
p. 012080, 2021. Available: 10.1088/1757-
899x/1022/1/012080 [Accessed 6 August 2022].

[17] S. Durga, R. Nag and E. Daniel, "Survey on Machine

Learning and Deep Learning Algorithms used in Internet of
Things (IoT) Healthcare," 2019 3rd International
Conference on Computing Methodologies and
Communication (ICCMC), 2019, pp. 1018-1022, doi:
10.1109/ICCMC.2019.8819806.

[18] M. Stoyanova, Y. Nikoloudakis, S. Panagiotakis, E.

Pallis and E. Markakis, "A Survey on the Internet of Things
(IoT) Forensics: Challenges, Approaches, and Open Issues",
IEEE Communications Surveys & Tutorials, vol. 22,
no. 2, pp. 1191-1221, 2020. Available:
10.1109/comst.2019.2962586 [Accessed 6 August 2022].

[19] J. Liu, "Business models based on IoT, AI and

blockchain", Master, Uppsala University, 2018.

[20] M. A. M. Ali, A. Basahr, M. R. Rabbani and Y. Abdulla,

"Transforming Business Decision Making with Internet of
Things (IoT) and Machine Learning (ML)," 2020
International Conference on Decision Aid Sciences and
Application (DASA), 2020, pp. 674-679, doi:
10.1109/DASA51403.2020.9317174.

[21] A. Israr, G. Abro, M. Sadiq Ali Khan, M. Farhan and S.

Bin Mohd Zulkifli, "Internet of Things (IoT)-Enabled
Unmanned Aerial Vehicles for the Inspection of
Construction Sites: A Vision and Future Directions",
Mathematical Problems in Engineering, vol. 2021, pp. 1-15,
2021. Available: 10.1155/2021/9931112 [Accessed 6
August 2022].

[22] F. Hussain, R. Hussain, S. A. Hassan and E. Hossain,

"Machine Learning in IoT Security: Current Solutions and
Future Challenges," in IEEE Communications Surveys &
Tutorials, vol. 22, no. 3, pp. 1686-1721, thirdquarter 2020,
doi: 10.1109/COMST.2020.2986444.

[23] J. Qiu, Q. Wu, G. Ding, Y. Xu and S. Feng, "A survey

of machine learning for big data processing", EURASIP J.
Adv. Signal Process., vol. 2016, pp. 67, May 2016.

59

View publication stats