Combining Multi-Agent Systems and

Arti cialIntelligence of Things: Technical Challenges

andGains
Pedro Luzolo

University of Technology of Belfort-Montbéliard

Zeina Elrawashdeh
ICAM, Site du Grand Paris Sud
Fatma Outay
Zayed University
Stephane Galland
University of Technology of Belfort-Montbéliard
Igor Tchappi
University of Luxembourg

Research Article

Keywords: Multiagent systems (MAS), Internet of things (IoT), Art cial intelligence (AI) and Smart
buildings

Posted Date: May 2nd, 2024

DOI: https://doi.org/10.21203/rs.3.rs-4293230/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Additional Declarations: No competing interests reported.

 Combining Multi-Agent Systems and Artificial
Intelligence of Things: Technical Challenges and
Gains
Pedro Hilario Luzolo1*, Zeina Elrawashdeh2,3 , Fatma Outay4 ,
Stéphane Galland1 , Igor Tchappi5
1* UTBM, CIAD UMR 7533, F-90010 Belfort, France.
2 ICAM, Site du Grand Paris Sud , F-77127, Lieusant, France.
3 Laboratoire Interdisciplinaire Carnot de Bourgogne, UMR 6303 CNRS

/ UTBM, F-90010 Belfort cedex, France.

4 College of Technological Innovation, Zayed University, Dubai, UAE.
5 AI-Robolab/ICR, University of Luxembourg, L-4365 Esch-sur-Alzette,

Luxembourg.

*Corresponding author(s). E-mail(s): [email protected];

Abstract
A Multiagent System (MAS) refers to a network of agents collaborating to
achieve same objective. This system comprises numerous individual programs
or hardware components (agents) that are simpler to construct and manage.
Additionally, these agents can dynamically and swiftly adapt to changes in
their environment. The MAS proves advantageous in addressing intricate issues
by employing the divide-and-conquer approach. It finds application in diverse
fields where the emphasis is on distributed computing and control, enabling the
development of resilient, adaptable, and scalable systems.
The Multiagent System (MAS) is not a substitute or rival for Artificial Intelli-
gence (AI) methods. Instead, AI techniques can be integrated within the agents
to enhance their computational and decision-making capabilities. The diversity
or uniformity of goals, actions, domain knowledge, sensor inputs, and outputs
among the agents in the MAS can determine whether each agent is heterogeneous
or homogeneous.
The Internet of Things (IoT) and Artificial Intelligence (AI) are two technologies
that have long been applied to the development of smart systems. These sys-
tems cover various areas, such as smart cities, energy management, autonomous
cars, etc. Intelligence, autonomy, and real-time monitoring are the fundamental

1
 elements that characterize these application areas. The convergence of artificial
intelligence (AI) and IoT, known as AIOT, Allows those electronic devices to
make intelligent, autonomous and more automatic decisions. This integration
leverages the power of MAS to enable intelligent communication and collabora-
tion among various entities, while IoT provides a vast network of interconnected
sensors and devices that collect and transmit real-time data. On the other hand,
AI algorithms process and analyze this data to derive valuable insights and
make informed decisions. The authors devoted efforts on the critical analysis of
AIOT research, highlighting specific areas with insufficient solutions and point-
ing out gaps for future advances. Essentially, the authors contribution is
in the formulation of innovative research directions, outlining a clear
guide for researchers and professionals in the expansion of knowledge
in AIOT integration. Research results in a significant contribution to
the continuous advance of the area, enriching the understanding of
challenges and boosting the development of solutions and strategies
in this technological convergence. Eleven research questions are considered
at the beginning of the review, including typical research topics and application
domains. From the SLR results the research directions are: (i) Development of
a methodology that shows how to integrate the different applications indepen-
dent of the scenarios that are deployed in. Additionally, elaboration of the tools
used in the integration process; (ii) Deployment of an agent in a microprocessor;
(iii) How to implement and connect Multi-agent systems (MAS) technology and
Internet of Things (IoT) devices (processors, controllers, sensors, and actuators).

Keywords: Multiagent systems(MAS),Internet of things (IoT), Artficial

intelligence(AI) and Smart buildings

1 Introduction
The implementation of Multi-Agent Systems (MAS) in the Internet of Things (IoT) is
crucial because it enables efficient coordination and adaptive decision-making in com-
plex environments. Agents facilitate interaction and collaboration among IoT devices,
optimizing operational efficiency, promoting intelligent automation, and providing
more robust and adaptable solutions across various domains such as smart cities,
healthcare, and industry. [1, 2] This integration extends the potential of IoT, allowing
it to handle the dynamics and heterogeneity of interconnected systems.[3]
A MAS is a system composed of multiple autonomous agents capable of com-
municating and working with each other, in order to achieve a common goal. MAS
is a suitable technology to handle distributed problem solving and decision making
[4].MAS focuses on intelligent agent behaviors and interactions to solve problems and
provide an improvement on technologies [4].
For example, MAS was applied in multiple research areas, such as transportation,
behavior modeling [5] and smart grids [6].
Internet of Things (IoT) objects are a typical example of a technology in a smart
building, in which devices could be considered as agents communicating in real time.
Indeed, according to Feng and Hu [7] and Fredj et al. [8], the IoT can be modeled as a

2
network of things or devices with embedded sensors, actuators, and software with the
purpose of connecting to other devices to exchange data through the Internet. What
can be gleaned from the article published by the Association of Computing Machinery
[9], To make IoT smart, one solution is to interconnect it with artificial intelligence
(AI) and MAS. In fact, agents should be able to apply autonomous behaviors when
associated with devices. According to Galster et al. [10], the synergy between IoT and
AI is known as Artificial Intelligence of Things (AIoT). IoT and AIoT are collecting
and using data [11–13]. Due to the limited processing capacity in IoT microcontrollers
that AI algorithms require. AI was first installed in the cloud and later on the edge.
However, cloud and edge computing present problems related to response speed
and information security response time; as sending data to the cloud or edge is usually
done with a certain latency. This latency can represent a security risk to the data, such
that on the path between the device and the cloud or edge, the data can be intercepted.
The time it takes to travel this path represents the duration of system processing or
latency. To make the system more distributed and mitigate data vulnerability, the
deployment of AI and agent-based modeling on IoT devices makes the system safe
and greatly improves latency [14].
MAS [1, 2] is a type of AI that can model and simulate complex systems that are
intrinsically distributed and composed of autonomous entities, called agents, that can
interact or communicate to achieve the goals of the global system. Therefore, DBOS
becomes a system composed of AIoT “agents” that are deployed on IoT devices. In
addition, MAS allows the safety and resilience of the system components, as well as
its scalability. [8, 15–17].
In this review of the literature, inspired by the systematic review of the literature
(SLR) method [18, 19], recent studies that employ MAS to represent AIoT are high-
lighted, as well as work that explores the application of MAS, for the development
of energy and daily activity management strategies. Methodologies, algorithms, and
frameworks are also extracted from the literature, as well as from the knowledge bases
and databases used in these studies. By synthesizing the results of this work, our SLR
aims to highlight the strengths and weaknesses of deploying MAS-based in IOT, dis-
cuss the challenges and opportunities associated with their use, and suggest research
direction for future research
This article is structured as follows. In Section 2, the general methodology applied
in this study is synthesized, as well as the main research questions that are at the heart
of this study. In Section 3, the results of the review are synthesized by answering the
research questions described above. Section 4 highlights the main research directions
extracted from the literature and the expertise of the authors. Section 5 is devoted to
other reviews that have been completed by the current study. Finally, the article is
concluded and several perspectives are provided.

2 Review Methodology
The authors [18, 19] describe Systematic Literature Review as a methodical and rig-
orous approach that aims to comprehensively collect, evaluate, and synthesize the
scientific publications available on a specific research topic. The fundamental principles

3
Fig. 1 Systematic literature review process, adapted from [10, 21, 22]

of SLR include a clear formulation of research questions, systematic search in scientific

and bibliographic databases, application of inclusion and exclusion criteria, evaluation
of the quality of selected articles, and structured summary of results. This approach
aims to minimize biases and provide a global and up-to-date view of knowledge in a
given field.
The original objective of the study, in the current SLR methodology is to provide
a research direction, to guide smart building decision-making and improve energy
consumption, so that the user comfort is achieved. Since then, SLR has spread to other
disciplines, including engineering sciences [20, 21], due to its ability, to rigorously and
transparently synthesize available knowledge.

2.1 Steps of the Methodology

The SLR methodology involves several key steps that are defined as follows:
1. Definition of research topics: In the context of a systematic literature review,
a research topic refers to the specific area of interest or focus that the review
aims to explore. It represents the subject or theme that the review addresses and
the research questions that it will seek to answer. A well-defined research topic
is crucial to conduct a systematic review of the literature, as it provides a clear
direction and scope for the study.

4
2. Definition of research questions: Research questions are specific inquiries or
interrogations that guide the focus and direction of a research study. They are
fundamental to the research process, as they help to define the scope, objectives,
and purpose of the investigation. Research questions are formulated to address
knowledge gaps, explore phenomena, investigate relationships, or solve problems
within a particular field of study. Formulating research questions often involves
a process of literature review, conceptualization, and refinement. They serve as a
road map for researchers, guiding their investigation, data collection, analysis, and
interpretation. Well-designed research questions lay the foundation for a rigorous
and meaningful research study.
3. Search engine selection and keyword definition: It is the process of iden-
tifying and defining the keywords or search terms that will be used to perform
the literature search. Keywords are specific terms or phrases that represent the
main concepts or themes related to the research topic. They are used to search
and retrieve relevant articles and publications from databases and other sources.
Keyword definition involves selecting and refining the most appropriate terms
that capture the essence of the research topic, and align with the research
questions or objectives of the review. It is a critical step in the systematic liter-
ature review process, as it determines the effectiveness and thoroughness of the
literature search.
4. Definition of selection criteria: Selection criteria are specific criteria or stan-
dards used to determine which studies are included or excluded from the review.
These criteria are applied during the screening and selection process, to ensure
that the chosen studies meet the predefined requirements and are aligned with
the objectives of the review.
Selection criteria are established based on the research questions, objectives,
and scope of the systematic review of the literature. They help maintain the
rigor, relevance, and quality of the review by ensuring that only studies that
meet certain criteria are included in the analysis. Keywords and criteria for the
inclusion and exclusion of scientific articles are defined in this step.
5. Coarsed selection of the articles: This point refers to the initial selection
process, where a large number of identified articles are screened at a high level,
to determine their potential relevance and suitability for inclusion in the review.
It is also known as an initial or preliminary screening.
During the coarse selection stage, the focus is on quickly assessing the articles
based on specific inclusion and exclusion criteria, to narrow down the pool of
studies for further evaluation. This initial screening is usually performed based
on the title and abstract of the articles, allowing a rapid assessment of their
relevance to the research topic.
6. Fine-grained selection of the articles: It is the more detailed and thorough
evaluation of the articles that have passed the initial coarse selection stage. During
the fine-grained selection process, the selected articles undergo a more in-depth
assessment to determine their suitability for inclusion in the review.
Fine-grained selection involves a comprehensive review of full-text articles,
rather than relying solely on the title and abstract. This stage aims to assess the

5
 articles against the predefined inclusion and exclusion criteria in a more detailed
manner, ensuring that only relevant and high-quality studies are included in the
final analysis.
The previously selected articles are analyzed in detail to extract answers to
the research questions and access the overall quality of each article. At the end
of this step, it is possible to synthesize the results and draw conclusions based on
the best available evidence and answer the research questions in a coherent and
rigorous manner.
The advantages of SLR lie in its rigor, exhaustiveness, and transparency, making
it possible to provide a global and up-to-date view of the state-of-the-art in a given
field. Through the use of impartial evaluation criteria and the elimination of selection
biases, a systematic literature review (SLR) provides a trustworthy and unbiased syn-
thesis of existing knowledge. However, SLR also has drawbacks, including the time
and resources required to perform comprehensive research and analysis of studies.
Furthermore, the results of an SLR can be limited by the quality and availability of
existing studies, as well as by the rapidly changing knowledge in some areas.
In the following sections, the different steps of the SLR methodology are applied
to the topics of this article.

2.2 Research Topic and Research Questions

The focus of the research and the specific inquiries or interrogations that guide the
research process in this document is the following: Identify and evaluate possible sci-
entific contributions to the applications and deployment of MAS in IoT devices. This
question is subbroken into several Research Questions (RQs) exploring key issues in
the context of the topics defined above. These RQ are formulated as follows:
• RQ1: What are the typical research topics and application domains that
are considered in the articles related to the application of AI in
IoT system? This RQ allows the researchers to know the topics of the
most discussed scientific fields and the related contributions, as well as the
applications that are considered for them.
• RQ2: What are the families of models or algorithms that are used for
designing or implementing AIoT or BOS? This question will provide
an overview of the approaches already applied in the field. Therefore, it will
be possible to determine the advantages and limitations of each approach.
• RQ3: What are the advantages and challenges related to the deployment
of AI model or algorithm on IoT devices? It allows one to highlight
the contribution of AI from the point of view of the authors of the reviewed
articles.
• RQ4: What are the technical characteristics or properties that have been
considered for implementing and deploying AI, and more specifi-
cally MAS, on IoT devices? Since most of the technology was deployed
in the cloud or on the edge, there is still an open question related to the
deployment of AI on the devices themselves. This RQ will help to under-
stand the constraints and limitations of AI software to be deployed in the
IoT. The question is split into the three following subquestions:

6
 a) What are the sensing and actuation characteristics to be considered as
a priority?
b) What are the run-time constraints and properties that MAS must
enforce for being deployed on IoT?
c) How to specify an MAS with run-time and deployment properties?

2.3 Search Engine Selection

The selection of article sources is a key point in the SLR methodology. Indeed, it
conditions all the rest of our study insofar as only the publications present on the
selected search engines are studied. The guidelines proposed by [23] have been used to
select search engines. Consequently, the following engines are selected: ACM Digital
Library, IEEE Xplore, and Google Scholar. The first two databases are the most used in
computer science. The third was selected because of its large number of documents that
are not indexed in the first two databases, for example articles from some conference
proceedings, masters, and Ph.D. theses. Although the indexation criteria used for the
publication of articles in Google Scholar are not the same as those used in the first
two databases, the topics discussed here are current and very useful.

2.4 Keyword Selection

Based on the author’s knowledge in the domains of computer sciences, especially MAS,
and electronic embedded systems, the topics of this review study, keywords are defined
as follows:
• AI: Since the application of AI in smart buildings, and especially DBOS, is a key
topic of this review study, the keywords in Eq. (1) are considered.
 
 artificial intelligence, 
ai = distributed artificial intelligence, (1)
multiagent system
 

• IoT: In a similar approach, the keywords related to IoT, and the associated
concepts are defined in Eq. (2).
 
iot, smart sensors,
iot = (2)
smart actuators, connected devices

• Extra Keywords: After a first screening of the results of the queries on the differ-
ent search engines, and based on the reviewers’ expertize, several articles where
missed. Consequently, to avoid the missing of articles in the reviewer process, the
keywords defined in Eq. 3 have been included.
 
instrumentation actuation technologies
extra = (3)
cyber physical systems

7
 The final search query is defined in Eq. (4), where the different classes of keywords
defined above are used.
ai ∪ iot ∪ extra (4)

2.5 Exclusion Criteria

The exclusion criteria are the predefined criteria used to determine which scientific
articles should be excluded from the review. These criteria help researchers filter out
studies that do not meet specific requirements or are not relevant to the research
objectives. The inclusion and exclusion criteria collectively shape the eligibility criteria
for study selection in a systematic review of the literature [10, 24, 25].
Exclusion criteria may vary depending on the research topic, objectives, and scope
of the review. Some common exclusion criteria include the following:
– ExC1: Not a recent research work. In order to use scientific articles with a
certain timeliness, the authors delimited the publication time of the arti-
cles to be consulted, thus establishing the consultation of articles with a
publication time equal to or less than 11 years. It means that articles with
a publication year < 2011, are excluded. It is assumed that the not recent
research is not up to date due to the high evolution rate of AI and IoT
technologies and usages.
– ExC2: Invalid type of article, the document is a poster or a demo. The
poster or demo is not considered consultation documents in this work,
because they are usually prepared very summarized and without many
details. It is assumed that a poster or a demo cannot give enough details of
the contributions, as there is not enough contributed content for evaluation.
– ExC3: Invalid type of article, the article is a survey. It is assumed that
the survey articles, that is, secondary studies, do not provide direct con-
tributions to AI models or AIoT technologies. These articles are briefly
discussed in Section 5.
– ExC4: Impossible to access the article text. Studies that cannot be accessed,
such as unpublished works or those with restricted access behind paywalls,
may be excluded from the review due to challenges in obtaining the com-
plete text or necessary data. Sometimes, it happens that the topic of a
certain scientific article and its summary are found in the database of a
certain scientific journal, but unfortunately, it is impossible to have access
to the complete content of the article.
It is impossible to evaluate an article when its text cannot be accessed
(PDF download, online text, etc.).
– ExC5: Duplicate study or extended article. The article is extended by
another article by the same authors. The contributions in the extended
article have been enclosed from those of the original article so that the
latter is excluded.
– ExC6: Unrelated to IoT. Articles whose content does not mention the use or
application of artificial intelligence or the Internet of Things cannot be used
as a bibliographic source for this work. These articles have not contributed
to the topics of IoT models or IoT technologies.

8
– ExC7: Unrelated to AI. Scientific articles whose content does not mention the
use or application of artificial intelligence cannot be used as a bibliographic
source for this work.
The article does not have a contribution to AI topics.

2.5.1 Quality Criteria

According to Brereton et al. [22], Calvaresi et al. [24], Galster et al. [10], Kitchenham
et al. [26], and Kitchenham [27], most SLRs use quality criteria to assess the quality
of primary studies. These criteria are based on the assessment on: (i) the context and
problems that are solved by the contributions, (ii) primary studies provide a sound
rationale for their contributions, (iii) results are clearly stated and discussed. Table 1
provides the four criteria that are used in the current review study.

Table 1 Criteria used for accessing the quality of the primary studies, adapted from [10]

# Description
Q1 Do the authors provide a sound rationale (i.e. motivation) for their work?
Q2 Is there an adequate description of the context in which the study has been conducted?
Q3 Is there a clear statement of the findings and the results including data that support the
findings?
Q4 Are the limitation of the study discussed and highlighted?

Note that, as is the case for Galster et al. [10], the quality criteria are not used
to exclude or include articles. Instead, they are used to report the overall quality of
primary studies, that is, articles that are not excluded from the SLR.

2.5.2 Biases and Disagreements

First, the review process assumed a common understanding among all reviewers about
search and analysis methods. Misunderstandings of concepts could potentially lead
to biased results. This threat was mitigated by providing the review protocol to all
reviewers and discussing it before the start of the review to ensure a selection process
as unbiased as possible.
To mitigate the subjectivity of the review process, certain measures were taken to
overcome biases and resolve conflicts. In particular, each selection step was conducted
by a principal reviewer, with an associated reviewer who may complete the analysis
of the first reviewer. A third reviewer intervened as a referee to resolve a conflict
in the exclusion/inclusion and in the research question-answering steps. Regarding
the quality assessment task, the quality assessments provided by the reviewers were
averaged for each article.

2.6 Methodology Application Summary

Because the number of articles provided by search engines is huge, especially by
Google Scholar, a stop criterion is applied [24, 28]: “Stop collecting articles only after

9
a sequence of 10 titles completely incoherent with the query that appeared in the list.”
It is the subjective responsibility of reviewers to determine the coherence/consistency
of an article when they believe that there is no link between the query performed in
the database and the title or abstract of the article that appears in the results.

Table 2 The number of articles through the different selection steps

Step Inputs Excluded Included

Initial query > 100000 - 427
Coarsed 427 318 109
Fine-grained 109 15 94

After applying the stop criterion described above, the total number of articles
selected directly from the search engines is 427 articles (see Table 2). The coarsed and
fine-grained selection steps allow to exclude unrelevant articles. 109 articles (25.5%)
have passed the coarsed selection, and finally 94 articles (22%) for the fine-grained
selection.

40

35 34

30 28
Number of articles

25

20

15
15

10 8

5 3
2 2
1 1
0
2014 2015 2016 2017 2018 2019 2020 2021 2022

Fig. 2 The number of articles per year after the selection steps

Fig. 2 provides the number of articles published in different years. It could be

observed that an increasing trend is observed in the number of articles published each
year. In particular, there were only a small number of articles published in previous
years (2013–2016), with the number gradually increasing over time. The number of
articles published in 2018, 2019, 2020, and 2021 is significantly higher compared to the
previous years, indicating a growing interest in the topic. The low number of articles
for 2022 could be explained by the fact that the search queries were done during
April 2022. Consequently, only articles published during the first part of 2022 have
been included in the study. Fig. 2 shows a growing trend in the amount of scientific

10
work carried out, from 2013 to 2021, with a slope of 2.833 ± 1.367. It confirms that
the interest of the scientific community in the application of AI in IoT has been
rapidly evolving. It is confirmed in a report by Markets & Markets [29] in which the
global IoT market size is expected to increase from $170.4 billion in 2017 to $561.0
billion by 2022, with a compound annual growth rate (CAGR) of 26.9% during the
forecast period. The report highlights the increasing role of AI in IoT applications, such
as predictive maintenance, energy management, and smart transportation. Another
report by McKinsey Global Institute [30] predicts that IoT and AI technologies will
generate significant economic value in the coming years, with the potential to create
up to $11.1 trillion dollars in economic value annually by 2025.

20
17
Number of articles

15
15
13

10

66
55
5 44
3333
222222222
11111111111111111
0
USA

Sweden

France
China
Greece
Spain

Austria
Brazil
Japan

Mauritius
India

Germany

Pakistan

Vietnam
Algeria

Croatia

Taiwan
Canada
Australia
Belgium

Italy

Russia

Bangladesh
Colombia

South Korea
Egypt
Estonia
Finland
Ireland

Sri Lanka
Korean Republic
Norway

Indonesia

Morocco
Netherland
Saudi Arabia
Thailand
United Kingdom

Portugal

Fig. 3 The geographical distribution of the authors of the selected articles per country and per
continent. An article may be counted multiple times if it is written by authors who are living in
different countries

The histogram in Fig. 3 provides information on the number of articles published

per country. USA, India and China are the three top countries in terms of the number
of articles published, with 17, 15, and 13 articles, respectively. This suggests that
these countries are likely to have a strong research focus in the area represented by
the articles. Furthermore, the remaining countries have published a relatively small
number of articles, with most countries publishing 6 articles or fewer. The map in
Fig. 3 shows the number of articles published per continent. Asia and Europe are
the continents with the highest number of articles published (48 and 44), followed by
North America (21). This suggests that there is a strong research focus on the topic in
these regions. The lower number of articles from Africa, Oceania and South America
may indicate that there is less research activity in these regions, or that authors from
these regions are less likely to be published in the particular journals or databases

11
included in the analysis. In general, this geographical distribution provides a useful
snapshot of research activity related to the topics of this study in different countries
and continents.

180 171

160
138
140
Number of articles

120

100

80

60

40
19
20
3 1 1 0
0
ExC7 ExC5 ExC3 ExC2 ExC4 ExC6 ExC1

Fig. 4 Number of articles that are excluded, per exclusion criterion

Fig. 4 shows the number of articles excluded based on the different exclusion criteria
that are explained in Section 2.5. It indicates that ExC7, which is used to exclude
articles that are unrelated to AI, has resulted in the exclusion of the highest number
of articles, i.e., 171 (51.4%). This suggests that a significant number of articles did
not contribute to the topics of AI. 138 articles (41.4%) are excluded because they
are duplicated in the results from the initial database query (ExC5). ExC3, which
is used to exclude survey articles, has resulted in the exclusion of 19 articles (5.7%).
ExC2, which excludes posters or demos that cannot provide enough details on the
contributions, has resulted in the exclusion of 3 articles. ExC5, which excludes articles
that are extended by another article by the same authors, has resulted in the exclusion
of 4 articles. One article is inaccessible (ExC4). One article is excluded (ExC6) because
it does not contain any contribution related to IoT. Finally, ExC1 has not resulted
in the exclusion of any articles. The use of exclusion criteria is an important step in
ensuring that the articles included in a study are relevant and meet certain quality
standards.

3 Results of the Review and Analysis

This section provides a detailed analysis of each RQ after the fine-grained selection
step. It should be noted that the results are related to the application of MAS in AIoT
in the context of smart buildings.

12
3.1 RQ1 - What are the typical research topics and
application domains that are considered in the articles
related to the application of AI in IoT system?
The number of articles per year is plotted in Fig. 2. However, to better understand
the evolution of MAS deployment in IoT devices for smart building applications in the
last decade, it is necessary to understand the related scientific and application topics.

3.1.1 Research Topics

A research topic refers to the subject or area of interest that a researcher chooses to
investigate and explore. It is the central idea around which the research is conducted
and provides a focus for the study. Table 3 provides the list of articles per research
topic. Fig. 5 shows the number of articles for each of these topics.

Table 3 List of articles per research topic

Research Topics Articles

T1 - Design, implementation of AI algorithms on the fog architecture [31]
T2 - Design, implementation of AI algorithms on the edge architecture [16, 17,
32–41]
T3 - Design, implementation of AI algorithms on the IoT devices [12, 42]
T4 - Design, implementation of IoT in industry 4.0 factories [7, 15, 43–
48]
T5 - Design, implementation of IoT in agriculture [49–56]
T6 - Design, implementation of IoT for improving processes in cities and smart cities [57–92]
T7 - Design, implementation of IoT for improving processes in health domain [12, 13,
93–99]
T8 - Increase efficiency, safety, and security of communications and devices in IoT [15, 100–
105]
T9 - Classification and characterization of concepts and properties in IoT environment [106]
T10 - Services and quality of services of AI [9, 107–
113]
T11 - Extract knowledge from large datasets [114]

The addressed research topics are listed below:

– T1: Design, implementation of AI algorithms on the fog architecture:
(1%) This research topic focuses on developing and implementing AI algo-
rithms on fog nodes or fog computing architecture. Fog nodes are intermediate
devices between the cloud and IoT devices that help reduce latency and
improve data processing efficiency. An example of this research topic would
be the design and implementation of an AI algorithm on a fog node to predict
equipment failure in an industrial setting.
– T2: Design, implementation of AI algorithms on the edge architecture:
(12.8%) This research topic focuses on designing and implementing AI algo-
rithms on edge devices or edge computing architecture. Edge devices are
located at the edge of the network and allow faster processing and reduced
data transfer. An example of this research topic would be the development of

13
 40
37
35

30

Number of articles
25

20

15
12
10 9
8 8 8
7
5
2
1 1 1
0
T6 T2 T7 T10 T4 T5 T8 T3 T1 T11 T9

Fig. 5 Number of articles per research topics

an AI algorithm for detecting anomalies in real time on a surveillance camera

located at the network edge.
– T3: Design, implementation of AI algorithms on the IoT devices: (2.1%)
This research topic focuses on designing and implementing AI algorithms
directly on IoT devices. By running AI algorithms on the device, real-time
decision making can occur without the need for communication with a cen-
tralized server. An example of this research topic would be the development
of an AI algorithm to predict soil moisture on a sensor located on a plant in
an agricultural setting.
– T4: Design, implementation of IoT in industry 4.0 factories: (8.5%) This
research topic focuses on developing and implementing IoT technologies in
industry 4.0 factories. Industry 4.0 refers to the fourth industrial revolution,
which involves the integration of physical and digital systems to increase
efficiency and productivity in manufacturing processes. An example of this
research topic would be the implementation of IoT sensors in a manufacturing
plant to monitor equipment status and predict maintenance needs.
– T5: Design, implementation of IoT in agriculture: (8.5%) This research
topic focuses on the development and implementation of IoT technologies
in agriculture. IoT can be used to monitor soil moisture, temperature, and
other environmental factors to optimize crop growth and reduce water use. An
example of this research topic would be the implementation of IoT sensors in a
greenhouse to monitor temperature and humidity levels and adjust irrigation
systems accordingly.
– T6: Design, implementation of IoT for improving processes in cities and
smart cities: (39.4%) This research topic focuses on developing and imple-
menting IoT technologies in cities to improve processes and make them more
efficient. For example, IoT sensors can be used to monitor traffic flow and
optimize traffic signals to reduce congestion. Another example is the use of

14
 IoT sensors to monitor air quality and adjust ventilation systems in buildings
accordingly.
– T7: Design, implementation of IoT for improving processes in health
domain: (9.6%) This research topic focuses on the development and imple-
mentation of IoT technologies in the healthcare domain to improve processes
and patient outcomes. For example, IoT sensors can be used to monitor vital
signs and alert healthcare providers if the condition of a patient deteriorates.
Another example is the use of IoT devices to monitor adherence to medications
and remind patients to take their medications.
– T8: Increase efficiency, safety, and security of communications and
devices in IoT: (7.5%) This research topic focuses on improving the effi-
ciency, safety, and security of IoT devices and communications. An example
of this research topic would be the development of secure communication
protocols for IoT devices to prevent unauthorized access.
– T9: Classification and characterization of concepts and properties in IoT
environment: (1%) This research topic focuses on developing methods to
classify and characterize concepts and properties in the IoT environment. An
example of this research topic would be the development of a classification
system for IoT devices based on their communication protocols and power
consumption.
– T10: Services and quality of services of AI: (8.5%) This research topic focuses
on developing and improving AI services in the IoT environment. It also
aims to enhance the quality of services offered by AI. Research on this topic
includes designing and developing novel algorithms, techniques, and models to
improve the performance, reliability, and security of AI-based services. It also
covers the design and implementation of AI-based services to cater to differ-
ent application domains in IoT. Examples include developing a framework for
evaluating the quality of AI-based services, designing intelligent algorithms
for resource allocation in IoT environments, developing techniques for load
balancing in AI-based services, and investigating the security and privacy
concerns of AI-based services. Another example is developing models to pre-
dict the performance of AI-based services under different conditions and to
evaluate their scalability in large-scale IoT deployments.
– T11: Extract knowledge from large datasets: (1%) This research topic focuses
on developing AI algorithms that can extract meaningful information and
knowledge from large and complex datasets generated in various IoT appli-
cations. With the increasing amount of data generated by IoT devices, it
becomes necessary to develop efficient algorithms to extract insights and pat-
terns that can aid decision-making and improve processes. An example of
this research topic is developing machine learning algorithms to analyze sen-
sor data from IoT devices in a manufacturing plant to identify patterns and
predict failures before they occur. Another example is using natural lan-
guage processing (NLP) techniques to analyze customer reviews and feedback
data from various sources to extract insights and trends related to customer
satisfaction and preferences.

15
 Fig. 5 shows the number of articles that correspond to each of the research topics
listed. It is evident that T6, which refers to the design and implementation of IoT
to improve processes in cities and smart cities, is the most widely covered topic in
the articles analyzed. T2, which pertains to the design and implementation of AI
algorithms on the edge architecture, has the second-largest number of articles, followed
by T7, which concerns the design and implementation of IoT to improve processes in
the health domain. T4, T5, and T8 each have 8 or 7 articles, which indicates that
they are also areas of active research interest. T10, which deals with the services and
quality of AI services, and T11, which concerns the extraction of knowledge from
large datasets. Each has 8 and 1 articles, respectively, suggesting that they are areas
of interest for some researchers but are not as widely studied as some of the other
topics. Finally, T1, T3, and T9 have the fewest number of articles, each with only one
or two articles, indicating that they are relatively less popular research topics. But it
still has some attention from researchers in the field. In general, the table gives a good
indication of current research interests and trends in the field of AI and IoT.

35
T1
T2
30 T3
T4
T5
25 T6
Number of articles

T7
T8
20 T9
T10
T11
15

10

0
2014 2015 2016 2017 2018 2019 2020 2021 2022

Fig. 6 Evolution of the numbers of articles published per year and per research topics

Fig. 6 shows the distribution of articles between the different research topics for
each year from 2014 to 2022, which can help researchers identify areas that require
further investigation. Research topics T6 (Design, implementation of IoT for improv-
ing processes in cities and smart cities) and T7 (Design, implementation of IoT for
improving processes in health domain) have consistently received more attention over
the years, with more articles published in these areas. T10 (Services and quality of
services of AI) has also gained attention, with an increase in the number of articles
published over the years. On the other hand, research topics T1 (Design, implemen-
tation of AI algorithms on the fog architecture), T2 (Design, implementation of AI
algorithms on the edge architecture), T3 (Design, implementation of AI algorithms on

16
the IoT devices), T8 (Increase efficiency, safety, and security of communications and
devices in IoT), T9 (Classification and characterization of concepts and properties in
IoT environment), and T11 (Extract knowledge from large datasets) have received less
attention, with only a few articles published in these areas.

3.1.2 Application Domains

RQ1 addresses the application domains of the articles. They refer to practical problems
that scientific and research studies tend to solve.

Table 4 List of articles per application domain

Application Domains Articles

D1 - Agriculture [39, 48–56]
D2 - E-health [12, 42, 94, 95, 97–99, 104]
D3 - Industry 4.0 [9, 15, 43, 44, 46, 47, 75, 86–88]
D4 - Mobility or energy management in cities [40, 57, 60, 63, 64, 66, 68, 69, 71, 74, 78–80, 82–
85, 89–92, 106]
D5 - Energy management in smart buildings [7, 41, 61, 65]
D6 - Security and safety in smart buildings [33, 59, 70, 101, 102, 114]
D7 - Multiple application in smart building [13, 16, 31, 32, 34–37, 45, 58, 62, 67, 72, 73, 76, 77,
81, 93, 96, 100, 103, 107, 109–113]
D8 - Big Data [38]
D9 - Cloud or edge computing [17, 105, 108]

30
27

25
23
Number of articles

20

15

11
10
10 9

6
5 4
3
1
0
D7 D4 D3 D1 D2 D6 D5 D9 D8

Fig. 7 Distribution of the articles per domain of application. Each article is associated to a single
domain

Table 4 provides an exhaustive list of articles per domain of application. And, Fig.
7 shows the number of articles per domain that are:

17
– D1: Agriculture: (10%) The application of AI and IoT in agriculture involves the
use of sensors and data analytics to improve crop yields, reduce waste, and
optimize resource usage. Examples of typical applications include precision
agriculture, which uses data from IoT sensors to create detailed maps of fields
and provide real-time information on soil moisture, nutrient levels, and crop
health; and livestock management, which uses IoT sensors to monitor animal
behavior and health.
– D2: E-health: (9.3%) The application of AI and IoT in e-health involves the use of
sensors and devices to monitor patient health and provide personalized care.
Typical applications include remote patient monitoring, which uses IoT sen-
sors to track vital signs and alert healthcare providers to potential problems;
and predictive analytics, which uses machine learning algorithms to analyze
patient data and provide early warning of potential health problems.
– D3: Industry 4.0: (11.5%) The application of AI and IoT in Industry 4.0 involves
the use of sensors and automation to optimize manufacturing processes and
improve product quality. Typical applications include predictive maintenance,
which uses IoT sensors to monitor equipment health and predict when main-
tenance is needed; and digital twins, which use AI algorithms to create virtual
models of physical systems and optimize their performance.
– D4: Mobility or energy management in cities: (25%) The application of
AI and IoT in energy and mobility management in cities involves the use
of sensors and data analytics to optimize transportation systems and reduce
energy consumption. Typical applications include intelligent traffic manage-
ment, which uses real-time data from IoT sensors to optimize traffic flow and
reduce congestion; and smart energy management, which uses IoT sensors to
monitor energy usage and optimize the distribution of energy across the grid.
– D5: Energy management in smart buildings: (5.2%) The application of AI
and IoT in energy management in smart buildings involves the use of sensors
and automation to optimize energy use and reduce costs. Typical applications
include demand response, which uses IoT sensors to adjust energy usage in real
time based on energy demand and pricing; and energy analytics, which uses
machine learning algorithms to analyze energy usage patterns and identify
opportunities for energy savings.
– D6: Security and safety in smart buildings: (6.2%) The application of AI and
IoT in security and safety in smart buildings involves the use of sensors and
automation to improve safety and security for building occupants. Typical
applications include intelligent video surveillance, which uses AI algorithms
to detect and respond to security threats in real-time; and smart fire safety,
which uses IoT sensors to detect fires and alert building occupants to evacuate.
– D7: Multiple application in smart building: (28.1%) The application of AI
and IoT in smart buildings involves the use of sensors and automation to
improve energy efficiency, optimize space usage, and improve occupant com-
fort. Typical applications include building automation, which uses IoT sensors
and smart controls to adjust lighting, heating, and cooling systems based on
occupancy and environmental conditions; and indoor air quality monitoring,

18
 which uses sensors to detect pollutants and allergens and provide real-time
feedback to building occupants.
– D8: Big Data: (1%) The application of AI and IoT in big data involves the use
of machine learning algorithms to analyze large data sets generated by IoT
sensors and devices. Typical applications include predictive analytics, which
uses historical data to make predictions about future trends or events; and
anomaly detection, which uses AI algorithms to identify abnormal patterns
or events in data that may indicate potential problems or threats.
– D9: Cloud or edge computing: (3.1%) The application of AI and IoT in cloud
or edge computing involves the use of distributed computing architectures
to process data generated by IoT sensors and devices. Typical applications
include real-time processing of sensor data at the edge, which allows for faster
response times and reduced latency; and cloud-based analytics, which allows
for the processing of large datasets and the deployment of machine learning
models.
Table 4 and Fig. 7 represent the distribution of articles among the different appli-
cation domains of AI and IoT. Based on the numbers, it appears that D7, which is
“Multiple application in smart building”, has the highest number of articles (27), fol-
lowed by D4, which is “Mobility or energy management in cities”, with 23 articles. On
the other hand, D8, which is “Big Data”, has the lowest number of articles with only
1 article, followed by D9, which is “Cloud or edge computing”, with only 3 articles.
It seems that the focus of research in this field is on multiple applications in smart
buildings and mobility or energy management in cities, indicating the importance of
these areas in the context of AI and IoT.

3.2 3.2 - What are the families of models or algorithms that

are used for designing or implementing AIoT or BOS?
This section discusses the simulation models that have been used in the reviewed
articles. Table 5 provides the list of articles for each type of model detected. Each

Table 5 List of articles per type of model

Type of Model Articles

M1 - Machine Learning [12, 16, 17, 32–38, 44, 46, 47, 49–54, 58, 59, 63–65,
67, 68, 70, 73, 75, 80, 81, 84, 93, 94, 98, 99, 112, 114]
M2 - Multiagent Systems [7, 12, 13, 17, 36, 38, 40, 42, 56, 60, 64, 66, 78, 102,
104, 107, 108, 111]
M3 - Computer Vision [16, 39, 48, 57, 59, 67, 72, 75, 77, 87, 95, 96, 113]
M4 - Knowledge Modelling [13, 15, 17, 32, 38, 43, 56, 64, 97]
M5 - Graph Theory or Queue Models [31, 83, 97, 101]
M6 - Mathematical Logic [9, 61, 76, 80, 82, 83, 86, 89, 97, 101]
M7 - Microservice-based Model or SaaS [15, 32, 39, 40, 43, 48, 51, 57, 62, 72, 77, 87, 95, 96,
109, 113]
M8 - General Object-Oriented Modelling [85]
M9 - Blockchain and communication protocols [12, 15, 55, 69, 78–80, 88, 91, 100, 101, 103–105]

19
 40 38

35

30

Number of articles
25

20 18
16
15
15 13
10
10 9

5 4
1
0
M1 M2 M7 M9 M3 M6 M4 M5 M8

Fig. 8 Distribution of the articles per family of models. Each article could be associated to one or
more families of models

type is defined in the following list:

– M1: Machine Learning. This family of models represents the family of machine
learning models that were used in the reviewed articles. Machine learning is
a field of study that focuses on developing algorithms and models that enable
machines to learn from data and make predictions or decisions without being
explicitly programmed. The following is a brief description of some of the
machine learning models used in the reviewed articles:
• Artificial Neural Networks (ANN): ANNs are computational mod-
els inspired by the structure and function of the human brain. They
consist of layers of interconnected nodes or neurons that process infor-
mation and make predictions. ANNs have been used in a variety of
applications, including image and speech recognition, natural language
processing, and predictive modeling.
• Convolutional Neural Networks (CNN): CNNs are a type of ANN
that are particularly well suited for image and video analysis. They use
convolutional layers to detect features in images and learn to recognize
patterns. CNNs have been used for tasks such as object detection and
classification, facial recognition, and autonomous driving.
• Deep Learning: Deep learning is a subset of machine learning that
involves training neural networks with many layers. Deep learning mod-
els have been used for a wide range of applications, including image and
speech recognition, natural language processing, and game play.
• Decision Trees (DT): DTs are a type of model that uses a tree-like
structure to make decisions. They are often used for classification and
regression tasks and can handle categorical and numerical data. Random
Forest is an ensemble learning method that combines multiple decision
trees to improve performance and reduce overfitting.

20
 • Q-Learning (QL): QL is a model-free reinforcement learning algorithm
that learns a policy (that is, a mapping from states to actions) by max-
imizing the expected cumulative reward. QL has been used in a variety
of applications, including robotics and game play.
• Clustering: Clustering is a technique used to group similar objects
together on the basis of their characteristics. Fuzzy C-Means and Fuzzy
K-Modes are two examples of clustering algorithms used in the reviewed
articles.
• Support Vector Machines (SVM): SVM is a type of supervised
learning algorithm used for classification and regression tasks. They find
the hyperplane that best separates the data into different classes. SVM
has been used in applications such as image and text classification and
speech recognition.
– M2: Multiagent Systems. This family refers to Multiagent Systems, which are
systems composed of multiple autonomous agents that interact with each
other and with the environment to achieve certain goals. The agents in the
system may have different capabilities, knowledge, and perspectives, and they
may collaborate, compete, or negotiate with each other to achieve common
goals. Multiagent systems can be used in various applications, such as smart
cities, transportation systems, supply chain management, and healthcare.
Some examples of Multiagent Systems used in the reviewed articles include:
• Coordinator Agent: A Multiagent System was proposed to optimize
the coordination between multiple stakeholders in a supply chain man-
agement system. The system included a coordinator agent who monitored
the supply chain and coordinated the activities of the other agents to
achieve the best possible outcomes.
• Reasoning and Belief Knowledge on graphs: Another Multiagent
System was proposed to represent and reason about beliefs and knowl-
edge in a social network. The system used graph theory to represent
the social network and the Neo4j tool to store and query the graph.
The agents in the system could infer beliefs and knowledge based on the
information available in the network.
• Autonomous Agents: A Multiagent System was proposed to control
traffic in a smart city. The system included autonomous agents that
could communicate with each other and with the traffic infrastructure to
optimize traffic flow and reduce congestion.
• AI-Privacy Agent: In a healthcare application, a Multiagent System
was proposed to manage the privacy of patient data. The system included
an AI-Privacy agent that used a decision tree algorithm to determine the
level of access to patient’s data for different stakeholders based on their
roles and permissions.
– M3: Computer Vision. Computer vision is a field of study that focuses on
enabling machines to interpret and understand visual information from the
world. Computer vision models are primarily based on machine learning
techniques and are designed to help machines recognize objects, scenes, and

21
 patterns in visual data. In the context of this study, computer vision models
can be used to analyze and interpret data from IoT sensors, cameras, and
other visual data sources to provide insight into the physical world and sup-
port decision-making processes. For example, computer vision models can be
used to monitor crop quality in agriculture, detect defects in manufacturing
processes, or track the movement of people and objects in a smart city.
– M4: Knowledge Modelling. This family of models is concerned with represent-
ing knowledge in a structured and formal way using ontologies. Ontologies
are explicit specifications of a conceptualization that define concepts and the
relationships between them. In the Internet of Things (IoT), ontologies can
be used to model the various devices, sensors, and data streams in a standard
way that enables interoperability and semantic understanding.
– M5: Graph Theory or Queue Models. This family refers to the use of graph
theory or queue models in IoT research. Graph theory is a mathematical disci-
pline that studies the properties of graphs, which are mathematical structures
that represent a set of objects and the connections between them. Queueing
models, on the other hand, are used to analyze and optimize queueing sys-
tems. In IoT research, graph theory is often used to model connected device
networks and to analyze the flow of information between them. For exam-
ple, a graph can be used to represent the connections between different IoT
sensors or devices, and algorithms can be applied to optimize the routing
of information through the network. Queueing models, on the other hand,
are used to analyze and optimize queuing systems in IoT environments. For
example, they can be used to optimize the scheduling of data transmissions
between IoT devices to minimize latency and maximize throughput.
– M6: Mathematical Logic. Mathematical logic is a subfield of mathematics and
deals with mathematical reasoning. This family of models uses mathematical
logic to represent and reason about knowledge and uncertainty. It includes
models such as logistic regression, linear temporal logic, fuzzy logic, and
modal logic.
• Logistic Regression Model: It is a statistical model that is used to
analyze and predict the relationship between variables. Logistic regres-
sion is widely used in machine learning for classification problems.
• Linear Temporal Logic (LTL): It is a formal logic that deals with the
reasoning about time and temporal relationships between events. LTL is
used in formal verification of hardware and software systems.
• Fuzzy Logic Model: It is a mathematical model that deals with reason-
ing that is approximate rather than precise. Fuzzy logic has been used
in various applications such as control systems, pattern recognition, and
decision making.
• Bootstrapping Algorithm: It is a statistical method that uses resam-
pling techniques to estimate the sampling distribution of an estimator.
Bootstrapping is used in many machine learning applications to estimate
the accuracy of a predictive model.

22
 • Modal Logics: It is a family of formal systems that extends classical
propositional and predicate logic to include operators expressing modali-
ties, such as necessity and possibility. Modal logic has been used in various
areas such as artificial intelligence, philosophy, linguistics, and computer
science.
These models are useful in applications that require reasoning about knowl-
edge and uncertainty, such as natural language processing, expert systems,
and decision-making systems.
– M7: Microservice-based Model or SaaS. Microservice architecture is a soft-
ware development approach that structures an application as a collection of
small, independent, and loosely coupled services, each serving a specific busi-
ness capability. Each microservice can be developed, tested, and deployed
independently, allowing greater flexibility and scalability. SaaS refers to the
delivery of software applications over the Internet as a service. Instead of
installing and maintaining software on a local computer or server, users access
the software through a web browser or mobile app. SaaS providers typi-
cally offer subscription-based pricing models in which users pay a monthly
or yearly fee for access to the software. In the context of the reviewed arti-
cles, the microservice-based model or SaaS was used to build and deploy
various types of applications, such as healthcare monitoring systems, chat-
bots, and recommendation systems. The microservices were often deployed
on cloud platforms, such as Amazon Web Services or Microsoft Azure, to
provide scalability and fault tolerance.
– M8: General Object-Oriented Modelling. This family refers to general
object-oriented modeling, where the primary modeling tool used is the Unified
Modeling Language (UML). UML is a standardized language used in soft-
ware engineering to create visual models of software systems. Overall, UML
provides a powerful modeling tool set to design, document, and communicate
the architecture and behavior of software systems.
– M9: Blockchain and communication protocols. This family of models
includes the use of blockchain technology, which is a decentralized and secure
way of storing and transferring data. Blockchain is a distributed ledger that
allows secure and transparent transactions without the need for a central
authority. Communication protocols are used to facilitate communication
between different parties in a blockchain network.
Fig. 8 shows the number of articles related to each family of models. It appears that M1
(Machine Learning) has the highest number of articles with 38, followed by M2 (Mul-
tiagent Systems) with 18 articles, and M7 (Microservice-based Model or SaaS) with 16
articles. M9 (Blockchain and communication protocols) has 15 articles. M3 (Computer
Vision) has 13 articles. M6 (Mathematical Logic) has 10 articles. M4 (Knowledge Mod-
elling) has 9 articles. M5 (Graph Theory or Queue Models) has only 4 articles. And,
M8 (General Object-Oriented Modelling) has the least number of articles with only 1.
The family of models cannot be determined from 7 articles [41, 45, 74, 90, 92, 106, 110].
It is important to note that the number of articles alone does not indicate the quality
or significance of the research conducted. It is possible that some families of models

23
Fig. 9 Scientific fields that are related or close to Artificial Intelligence. This figure is the fusion of
those which are proposed by Osamy et al. [115], Cao et al. [116], Kaveh and Mesgari [117]

have fewer articles, but they are more impactful and influential in the field. Further-
more, the research topics of the articles and the quality of the studies should also be
considered when evaluating the contributions of each family of models to the field.
The different families of models are used in different ways to solve different types of
problems in artificial intelligence, machine learning, deep learning, and data science.
While there may be some overlap between the families of models, each family has its
own unique characteristics and applications, as illustrated in Fig. 9.

3.3 RQ3 - What are the advantages and challenges related to

the deployment of AI model or algorithm on IoT devices?
To provide an analysis of the advantages of AI deployment in IoT devices, it is
important to understand how and where AI is deployed. On the basis of the answer
related to 3.2, there are three levels of processing data, as seen in Fig. 10 and
defined by Buyya et al. [118], Gubbi et al. [119], Bonomi et al. [120], Armbrust et al.
[121], Satyanarayanan et al. [122].
• Cloud Computing: It refers to the practice of using a network of remote servers
hosted on the internet to store, manage, and process data, rather than a local
server or a personal computer.
• Fog Computing: It is a distributed computing paradigm that aims to bring com-
puting power and storage capabilities closer to the network edge, where data are
being generated and processed, rather than relying solely on the cloud for data
processing and analysis.
• Edge Computing: It refers to the practice of deploying computing resources and
capabilities directly on IoT devices, such as sensors, wearables, and smart devices,
rather than relying on the cloud or a central server for data processing and
analysis.
Deploying MAS on IoT devices is related to edge computing. Indeed, agents are
autonomous entities that are capable of interaction to reach a common goal in the
system [1, 2]. When the IoT devices are assimilated to agents, it becomes possible to
make them autonomous, reactive, proactive, and interact with the other IoT devices.
From the reviewed articles mentioned in Table 6, two main reasons for supporting

24
Fig. 10 Different layers of the IoT and data management system composed by the cloud, fog and
edge layers, initially proposed by Alam et al. [123]

Table 6 List of articles which are related to, at least, one layer of the IoT and data management
system, as illustrated in Fig. 10

Layers in the IoT systems Articles Count

Cloud layer [7, 9, 12, 13, 33, 37, 39, 45, 46, 48, 52–54, 60–62, 66, 68, 69, 41
73, 77, 78, 82–84, 86–88, 92, 94, 96, 105, 106, 110, 112]
Fog layer [15, 31, 42, 43, 50, 57, 64, 71, 72, 74, 89–91, 93, 97, 99, 102, 113] 34
Edge layer [17, 34, 36, 38, 40, 44, 47, 51, 55, 56, 58, 63, 65, 67, 70, 76, 80, 32
85, 95, 98, 100, 104, 114]
Not specified [16, 32, 35, 41, 49, 59, 75, 79, 81, 101, 103, 107–109, 111] 15

the choice of edge computing could be highlighted. First, the idea shared by the arti-
cles that used fog computing aims to complement and not replace cloud computing
through the use of an intelligent router or gateway in order to mitigate the amount of
data sent to the cloud. Second, the idea shared by the articles that uses edge comput-
ing is based on creating a scalable, redundant, low latency, integrable network with
new and old devices, reducing network usage, greater data security, and versatility.
The above analysis enables one to give an answer to RQ3 after some refinement.
Therefore, the main advantages of deploying AI in IoT devices, i.e. considering edge
computing, are synthesized below. The distribution of the reviewed articles on these

25
advantages is shown in Table 7. An article may mention multiple advantages. There-
fore, it is counted multiple times in this table. The advantages that the application

Table 7 List of articles per mentioned advantage of AI in IoT devices

Advantages Articles Count

AD1 - Accuracy of data and behaviors [7, 9, 12, 13, 15, 17, 32, 35, 37–39, 42– 61
45, 47, 49–54, 56–63, 65–69, 71, 74–
79, 81–83, 85, 87–90, 93, 95, 96, 106–
109, 112, 114]
AD2 - Safety of IoT devices and quality of service [12, 13, 15, 17, 32, 37, 38, 43–45, 47, 34
49–51, 57–60, 65–69, 84–86, 88, 91–
93, 95, 96, 111]
AD3 - Security inside and between nodes [7, 33, 36, 46, 63, 69, 70, 74, 75, 77– 21
80, 82, 89, 91, 104, 106, 108, 112, 113]
AD4 - Efficiency of the AIoT nodes [12, 13, 15, 17, 31, 34, 36–40, 43, 44, 40
47, 49–51, 55, 57–59, 64, 65, 67–69, 72,
76, 85, 87, 92, 93, 95, 96, 98–101, 103]
AD5 - Green computing [7, 12, 13, 15–17, 37, 38, 40, 43, 44, 46, 39
47, 49–51, 57–59, 63, 65, 67–69, 72, 74,
75, 77, 79, 82, 86, 89, 93, 95, 96, 100,
105, 106]

of AI in IoT devices bring are enumerated below:

– AD1: Accuracy of data and behaviors. By deploying AI models on edge
devices, the margin of error can be reduced because the models can oper-
ate closer to the data source and end users of the application. This enables
real-time processing of data and provides more accurate predictions and
insights. When processing data on the edge, the amount of data that needs
to be transferred over the network is reduced, which minimizes latency and
improves the responsiveness of the system. Furthermore, the use of edge
computing can improve the reliability and consistency of data processing
and analysis, as the devices can operate even when there is limited or no
network connectivity. Finally, AI on the edge can enhance the accuracy and
reliability of data processing and analysis by bringing computation closer
to end users. This can help improve the performance and efficiency of the
system and enable real-time insights and predictions.
– AD2: Safety of IoT devices and quality of service. Edge computing can help
improve the safety and quality of IoT device service by reducing network
usage and improving connectivity between systems and applications. This
can help anticipate failures and apply redundancy to ensure that devices
operate optimally. Additionally, the use of AI on edge devices can enable
rapid intervention when problems occur, leading to faster resolution of
issues.
– AD3: Security inside and between nodes. Edge computing can help improve
IoT device security by processing and storing data locally, rather than trans-
ferring it to external servers. This reduces the need to secure communication

26
 using cybersecurity mechanisms and enables GDPR compliance by keeping
personal data within the device. By storing data in multiple locations, criti-
cal information can be filtered and processed locally, while only nonsensitive
information is sent externally, improving security and privacy.
– AD4: Efficiency of the AIoT nodes. Deploying AI on IoT devices enables ser-
vices to be delivered quickly, with lower latency and reduced cost compared
to using cloud-based solutions. The use of edge computing enables AI mod-
els to be executed locally, making the path that data takes much shorter,
leading to faster response times. This can help achieve quick and easy access
to services, and make various applications run more efficiently.
– AD5: Green computing. The use of edge computing and AI can help optimize
resource and energy management, leading to more sustainable computing
practices. This is because edge devices can reduce the need for data to be
transferred over long distances, which can lead to lower energy consump-
tion and carbon emissions. Additionally, by enabling devices to process and
store data locally, less energy is required to maintain communication with
external servers.
Among all reviewed articles, 7 articles have not provided arguments related to the
advantage of edge computing [41, 48, 73, 94, 97, 102, 110]. From Table 7, it can be
seen that the reviewed articles have mentioned AD1 (61 articles), followed by AD4
(40 articles), AD5 (39 articles), AD2 (34 articles) and AD3 (21 articles) as advantages
of edge computing. This suggests that the accuracy of data and behaviors is the most
widely discussed advantage of using edge computing and AI in devices, followed by
the efficiency of AIoT nodes and green computing. The safety of IoT devices and the
quality of service are discussed less frequently, and security within and between nodes
is discussed least. However, it should be noted that the number of references alone
does not necessarily indicate the importance or validity of each advantage. However,
despite these important advantages, the literature has also revealed some challenges
and barriers due to the combination between artificial intelligence and the Internet
of Things. These challenges can be divided into two categories; some of them are
technological barriers, while others are more social. These two categories are presented
in the following two subsections.

3.3.1 Technological challenges

Technology is always accompanied by opportunities and barriers. The study presented
in [124] has focused on several challenges, such as:
1. Cybersecurity and data risks: The domain of smart applications has networks
that generate a large amount of data. Sharing these data raises many challenges;
as this data might be private, which can be a threat for smart city systems.
2. Smart waste and hygiene management: Handling waste is a major chal-
lenge in most countries. AI with IoT should find solutions to collection,
transportation,treatment, recycling, and disposal in waste management.
3. Managing energy demands: the energy demand and costs are increasing over
time. Domestic energy consumption is exponentially increasing due to the use of

27
 modern televisions, air conditioners, washing machines, smartphones, and com-
puters. Technological advances and changes in consumer habits lead to higher
energy demands, and now energy producers are looking for help from AI and the
Internet of Things to optimize the distribution of energy demands through meth-
ods such as automating streetlights, increasing unit electricity prices during peak
times, and updating old equipment with modern equipment.

3.3.2 Deployment and Societal Challenges

1. Lack of professionals: The adoption of AI and the IoT require highly skilled
professionals. Without adequate experience and knowledge, organizations always
misunderstand the benefits of these technologies. One of the current challenges
facing smart cities is the lack of professionals with knowledge of computer
technology and knowledge of different fields.
2. Lack of Trust in AI and IoT: The lack of trust in AI and IoT could slow their
implementation in smart city development. As mentioned by researchers, building
trust AI models that can transform social, political and business environments
and help people in decision-making processes will remove any negative opinions
about the usage of AI and IoT technologies.
3. Insufficient Funds or Capital: AI and IoT techniques should help prioritize
projects according to severity, requirements, timelines, and other parameters, and
optimally allocate budgets using the best optimization techniques. To effectively
address this problem, the government may need to obtain additional funding from
entities in the private sector that are interested in these smart city projects.

3.4 RQ4 - What are the technical characteristics or properties

that have been considered for implementing and deploying
AI, and more specifically MAS, on IoT devices?
To answer RQ4, the technical characteristics and properties that have been taken
into account when implementing and deploying AI, and specifically MAS, on IoT
devices must be studied. In recent years, there has been a growing interest in devel-
oping intelligent systems that can operate in the IoT environment, as this can enable
a wide range of applications that can benefit from the capabilities of both AI and
IoT. However, implementing and deploying AI on IoT devices presents a number of
technical challenges, including limited processing power, memory and storage capac-
ity, network bandwidth, and power consumption. Therefore, researchers have explored
various approaches to address these challenges and enable efficient and effective AI
deployment on IoT devices. In this section, some of the technical characteristics and
properties that have been considered in this context are explored.
The answer to RQ4 is split into different subquestions:
• What are the types of sensors and actuators used in edge computing applications?
• What types of controllers and computing resources are used to deploy AI in edge
computing applications?

28
3.4.1 Sensors and Actuators
The sensors are selected according to the type of variable to be read (temperature,
presence, brightness, speed, weight, etc.). According to the type of variable chosen, the
sensor can be digital or analog. The analog sensor is used to read variables that do not
change abruptly as a function of time (temperature, pressure, speed, etc.). The digital
sensor is used to take readings that vary sharply (presence, length, weight, etc.).

Table 8 List of sensors or actuators that are considered for deloying AI on IoT

Sensors and Actuators Numbers of Articles

Camera (RGB or thermal) 75
Temperature sensor 70
Humidity sensor 69
Gas sensor 40
Presence sensor 40
Ultrasound sensor 37
Water flow sensor 36
Servo-motor actuator 47
Relay 35

Table 8 presents a list of sensors and actuators that have been considered in the
reviewed articles. The most popular choices among the articles are camera with a count
of 75, followed by temperature sensor with 70, humidity sensor with 69, and gas sensor
and presence sensor both with 40 counts. Other technical choices include ultrasound
sensor with 37, water flow sensor with 36, servo-motor actuator with 47, and relay
with 35. These choices illustrate the variety of options available for implementing and
deploying AIoT.
Finally, two factors are important to be considered when implementing an AIoT
application: (i) the variables to be measured, and (ii) speed of variation of each
variable.

3.4.2 Controller and Computing Resources

Choosing the right type of controller is an important decision when deploying AI
on IoT devices. The two most common options for controllers are Raspberry Pi and
Arduino. Both have advantages and disadvantages, and the choice will depend on the
project’s specific needs.
Raspberry Pi is a popular choice because of its processing power and flexibility. It
can handle complex AI algorithms and can be easily programmed using languages such
as Python. Additionally, it has built-in WiFi and Bluetooth connectivity, making it
easy to connect to other devices and networks. However, it’s high power consumption
and cost may not be ideal for certain IoT projects.
On the other hand, Arduino is a low-cost and low-power option that is suitable
for simple AI applications. It is easy to use and program, and it can be powered by
batteries or solar cells. However, it has limited processing power and memory, which
may not be sufficient for more complex AI algorithms.

29
 ESP is another popular hardware controller that is used to deploy AI on IoT
devices. It is a low-cost, low-power microcontroller that is designed for WiFi con-
nectivity and is ideal for IoT applications that require remote monitoring and
control.
Ultimately, the choice between these controllers will depend on the specific require-
ments of the AIoT system, including the desired level of processing power, energy
consumption, and cost. It is important to carefully evaluate the options and choose
the controller that best meets the needs of the project.
This point of view is shared by Lv et al. [69] and Prutyanov et al. [16], and could
be synthesized in the advantages and disadvantages of both types of controllers in
Table 9.

Table 9 Advantages and disadvantages of the Rasberry PI and Arduino controllers for deploying
AIoT on the edge

Raspberry PI Arduino
Advantages Disadvantages Advantages Disadvantages
Greater processing Need for extra periph- Easy to connect sen- Single task
power erals sors and components
More connectivity Boot required Wide variety of shields Low speed
More languages and Needs installation and Cheaper Lack of native connec-
easy configuration tion
Multi-tasking It can get expensive Don’t need a lot of Great C++ learning
equipment curve

Raspberry is slightly larger than Arduino. It is a microprocessor (SBC-Single

Border Computer) that processes programs and has an embedded microcontroller.
Arduino is a microcontroller, controls components; It is an open-source project [16, 69].
After fine granulation, the three most used microcontrollers / processors in the
scientific articles of this phase are highlighted in Table 10. Certain works are identified,
although specifying the type of sounders they use, depending on the specific objectives
of the work, but do not specify the type of controller/processor used, these works in the
graph are identified as IoT. As can be seen in Table 10, the raspberry pi microprocessor
is the most used with 29%, of a total of scientific articles after fine granulation. Out of

Table 10 Most Used Micro-Processor/Controller for AIOT Solutions

Controllers Articles
Raspberry PI [12, 16, 17, 32, 44, 49, 52, 53, 55, 56, 58, 63, 64, 67, 76, 84, 90, 93, 98, 99, 111]
Arduino [12, 13, 39, 42, 48, 49, 52–56, 59, 64, 67, 85–87, 90, 92, 98, 111]
ESP [13, 39, 42, 47, 50, 51, 54, 57, 85, 87, 92, 103]

the 94 articles surveyed, 58 (61.05%) do not mention the type of controller to be used
for deploying AI on IoT devices. This indicates a lack of emphasis on the importance
of choosing the appropriate hardware controller for the intended use case. While it

30
may be possible to deploy AI on various hardware platforms, including Raspberry Pi,
Arduino, and ESP, each has its strengths and limitations. Choosing the appropriate
hardware controller is important to ensure optimal performance and efficiency for
the intended application. Therefore, it is crucial that future research in this area
emphasizes the importance of selecting the appropriate hardware controller for the
intended use case. Among the 36 articles that a controller is mentioned (see Table 10),
it appears that an equal number of articles (21 each) mention using Raspberry Pi or
Arduino. Furthermore, the articles in 12 mention the use of the ESP controller.

3.5 Quality Assessment

Based on the quality criteria defined in Section 2.5.1, the quality of the articles was
evaluated according to the four quality criteria defined in Table 1. They are, respec-
tively, related to the explanations of the motivations (Q1), the study context (Q2), the
theoretical and experimental results (Q3), and the limitations and research directions
(Q4). Fig. 11 shows the average evaluation for each of these criteria. The reviewers
have provided a score, based on their background and knowledge, based on three lev-
els of quality: “insufficient” (score: 0), “good” (score: 2), “excellent” (score: 4). The
numerical values have been selected with a step of 2 to allow a reviewer to select a
score in the middle if he/she hesitates between two adjacent quality levels. Each article
was evaluated by at least two reviewers, and the results are averaged.

4.5
4
4

3.5
3 3
3
Quality Score

2.5
2
2

1.5

0.5

0
Q1 Q2 Q3 Q4

Fig. 11 Quality assessment for the reviewed articles in the range [0; 4]

For Q1, which evaluates whether the authors provide a solid rationale for their
work, the articles receive a score of 4, which indicates an excellent evaluation. This
suggests that the authors effectively provided a clear justification for their research
and why it is important. For Q2, which assesses whether there is an adequate descrip-
tion of the context in which the study was conducted, the articles obtain a score of

31
3, which indicates a good evaluation. This suggests that although the authors pro-
vided a description of the study context, there may have been some areas where more
detail could have been provided. For Q3, which evaluates whether there is a clear
statement of the findings and the results, including data supporting the findings, the
articles receive a score of 3, which indicates a good evaluation. This suggests that the
authors provided clear and supported findings from their research. Finally, for Q4,
which assesses whether the limitations of the study were discussed and highlighted, the
articles receive a score of 2, which indicates a fair evaluation. This suggests that while
some limitations of the study are discussed, there may have been areas where more
limitations could have been highlighted, or several articles do not mention limitations
and future research direction.

3.6 Summary
we highlight that as the typical research topic, we have the two topics most explored
by researchers:
• Design, Implementation of IoT for Improving Processes in Cities and
Smart Cities: with 37 themes.
• Design, Implementation of Ai Algorithms on the Edge Architecture:
with a total of 12 themes.
The domain of applying these research is more linked to ”Multiple Application
in Smart Building”, the most utilized AI model type is Machine Learning and
then is the multi-agent system.
Regarding the disadvantages and challenges related to the deployment of Mult-
Agent systems in IoT devices, of the three levels of data processing (Cloud, Fog
and Edge Computing), Edge Computing showed in advantage for the following
reasons: Allows SMART IOT DEVICE , low latent, reduces the cost of con-
nectivity, improves safety and privacy in processing and data analysis on the
device, general deformation improves the system: autonomy, responsibility and
adaptation.
Regarding the technical properties considered for the implementation of arti-
ficial intelligence or multi-agent systems on IoT devices, we have to say that the
Micro-Processors (Raspbarry Pi) are better from micro-controllers (arduino), this
is due to:
• Greater processing power;
• More connectivity and More languages;
• Easy Multi-tasking.

4 Main Research Directions

According to our quality analysis, which is described in Section 3.5, numerous reviewed
articles do not provide explicit limitations and perspectives for their research. Based
on the research work performed and the expertise of the authors, five major research
directions could be highlighted.

32
4.1 Author’s Contribution
In this systematic literature review, the Scientific articles revised by the authors reveal
significant gaps. From the Subsection 2.2 and addressed in section 3 we can highlight:
• Accuracy of Data: Ensuring the accuracy and reliability of the data processed
at the edge is crucial for effective decision-making. Noisy or inaccurate data can
lead to sub-optimal performance of AI models.
• Parameter Updates: Efficient mechanisms for updating model parameters
while considering bandwidth constraints and real-time requirements need to be
developed.
• Energy Efficiency in Industrial IoT Applications: optimization of energy
consumption in complex industrial settings. Developing AI-driven solutions that
effectively manage energy usage in industrial environments, considering diverse
and dynamic operational requirements, remains a significant challenge.
• Integration of Green Computing with Smart Energy and Power
Systems Modeling: develop comprehensive frameworks and algorithms that
consider the dynamic nature of energy sources, demand patterns, and the
heterogeneous nature of IoT devices.
The gaps identified by the authors constitute crucial elements that underlie the
research directions outlined in the subsequent paragraphs of this scientific article.
The Careful identification of these gaps during the bibliographic review not only
justifies the need for additional research, but also establishes a fundamental basis for
the formulation of research strategies outlined in the subsequent paragraphs of this
scientific work.

4.1.1 Edge AI for increasing autonomy of IoT systems

The integration of Edge AI into IoT systems has the potential to greatly increase
their autonomy, allowing intelligent and autonomous deployments. However, this also
presents a number of challenges, such as the storage of training data, model training,
and parameter updates. To address these challenges, research is being conducted on
how to make a smart home more autonomous using machine learning applications
deployed on edge servers. Service accuracy is used as a metric in the offload of mobile
edge computing (MEC) traffic [44]. Furthermore, different deep neural networks such
as CNN (Conventional Neural Network), RNN (Recurrent Neural Network), etc. are
being considered and compared with existing techniques [33].
In order to extend the architecture to allow collaboration between multiple edge
clusters and dynamic repartition of control, researchers have proposed an extended
architecture [17]. Additionally, new lossless biosignal steganography approaches are
being developed with a higher data hiding capacity, which will be implemented on
wearable devices [38]. To ensure precision, a methodology for the conversion of PET
scan images to voxels is being developed, enabling the creation of a 3D representation
and visualization of tumors [12].
Other areas of research in Edge AI for IoT include integrating image compres-
sion algorithms in IoT data from agriculture and agricultural industries, modeling
collaborative and distributed heterogeneous operating systems, and deploying AI on

33
devices in a smart home context [55, 79, 80]. There is also an interest in perform-
ing experiments on real fish farms, improving classification accuracies and deploying
the model on ESP32 while integrating connectivity media such as LoRaWan and SIM
Card Router [39].
In general, the integration of edge AI into IoT systems has a great potential to
increase autonomy and intelligence in a variety of domains. However, challenges such
as accurate data, dynamic repartition of control, and model training and updates need
to be addressed through continuous research and development [40, 111].

4.1.2 Green AIoT

The Green AIoT research direction aims to explore how to make the use of AI and
IoT more environmentally friendly and sustainable. This area of research is becom-
ing increasingly important as the energy consumption of connected devices and data
centers increases, leading to higher carbon emissions. Several studies have proposed
ways to address this issue by applying AI to energy management in homes and indus-
tries. For example, SmartME, a smart home battery proposed by Bruneo et al. [57],
uses AI to optimize energy storage and distribution, reducing overall energy consump-
tion. Similarly, the study by Hazarika et al. [60], Manman et al. [66] suggests the
use of Cognitive Radio Sensor Networks (CRSNs) for energy-efficient industrial IoT
applications.
Other research directions in the Green AIoT domain include investigating the
links between Green Computing and Smart Energy and Power Systems Modeling, as
described in [61]. This study identifies research and innovation opportunities in energy
information technology to address the IoT energy scarcity problem and proposes the
use of AI to optimize energy consumption in homes and buildings. Additionally, some
studies focus on the application of AI to make cities more sustainable, such as Smart
City policies proposed in [79].
One critical aspect of green AIoT is water quality monitoring. Studies such as Lak-
shmikantha et al. [85] propose the use of the latest sensors and IoT technologies for
better monitoring of water quality, enabling the implementation of better water man-
agement policies. Furthermore, this study suggests the use of wireless communication
standards to improve communication between sensors and the cloud, further reducing
energy consumption.
In conclusion, the direction of the green AIoT research direction aims to make the
use of AI and IoT more sustainable by optimizing energy consumption and reducing
carbon emissions. Several proposals have been made, such as SmartME, the use of
CRSN, and Smart City policies. Water quality monitoring is another critical aspect
of Green AIoT, and several studies propose using the latest sensors and wireless com-
munication standards for better water quality monitoring. By pursuing these research
directions, we can ensure that AIoT growth is sustainable and eco-friendly, benefiting
both society and the environment.

4.1.3 Ethics and Human-in-the-loop

The third research direction for the future of AIoT is focused on ethics and the role of
humans in the loop. With the increasing adoption of AI and related technologies, it is

34
essential to ensure that these systems are designed with the user in mind, promoting a
more user-centric approach. Several studies have suggested the use of AI as a modern-
ization method to improve user experience, especially in the context of smart homes
and farming. For example, research has proposed a mobile system that can help at
every stage of agriculture, from seed sowing to harvesting crops [49].
However, the implementation of AI in various ecosystems must consider ethical
implications, including the potential loss of resilience in living systems such as agricul-
tural landscapes and forest ecosystems [35]. To ensure trust and confidence in AIoT
systems, more experiments with users are required to strengthen their confidence in
the capture and representation of smartness [9]. Researchers have also proposed an
ethical design for IoT, including the incorporation of user, usage, and context infor-
mation through a unified semantic representation, driving an adaptation mechanism
with the aim of providing personalized services and optimizing user experience [89].
As AI and associated technologies continue to automate various tasks, it is neces-
sary to consider the role of humans in the loop and the acceptability of AI in their
daily lives. Therefore, XAI (Explainable Artificial Intelligence) has been suggested as
a means to increase the acceptability of AI in the smart home [125]. The integration
of robotics, smart sensor technology, and AI systems is also proposed to automate the
entire farming procedure, from seeding to intelligent fruit and vegetable harvesting
and packaging [53].
In conclusion, the ethical and human-centric design of AIoT systems is critical
to ensure their acceptance, trustworthiness, and effectiveness in various ecosystems.
Future research in this direction should focus on designing AI systems that are more
explainable and transparent to end users and integrating user, usage, and context
information to provide personalized and optimized services.

4.1.4 Security and Privacy in AIoT

As AIoT becomes more prevalent in various domains, security and privacy concerns
become increasingly important. There is a need to develop robust security measures
to protect data and ensure the privacy of users. Several research directions have been
proposed to address this issue.
One direction is to examine the tunneled session by obtaining the key in the secu-
rity analysis of the key distribution step and understanding the encryption algorithm
through the decompilation of the android package [62]. This approach aims to iden-
tify vulnerabilities in the encryption algorithm and the key distribution process to
improve security.
Another promising direction is to explore the use of blockchain as a service, trust
management, and security by design approach [58]. Blockchain technology can provide
a decentralized and tamperproof method of managing transactions and data. By using
blockchain, trust management can be improved, and security can be built into the
design of the AIoT system.
In addition, encryption on the communication link and optimizing the current
management system, such as load balance for data read and stored, can improve system
security [58]. This approach focuses on ensuring the confidentiality and integrity of
data transmission between devices and storage systems.

35
 Furthermore, a study on the security of the Advanced Metering Infrastructure
(AMI), which is a key element of the Smart Grid, is needed. The study will focus on
how to securely integrate energy-aware smart homes, equipped with smart meters and
smart appliances [91]. This approach aims to enhance the security of the Smart Grid,
which is critical for maintaining the stability and reliability of the power grid.
Finally, there is a need to model systems with an extended architecture that allows
collaboration between many edge devices and dynamic loading with security [91]. This
approach focuses on designing AIoT systems that can adapt to the changing needs of
users while ensuring security and privacy.
In conclusion, as AIoT continues to grow and evolve, it is crucial to consider the
security and privacy of these systems. The research directions mentioned above provide
a starting point for developing robust security measures for AIoT systems.

4.1.5 Safety and Performance of Communication and Computing

Resources in AIoT
As AI and IoT continue to grow and integrate into our daily lives, it becomes crucial
to ensure that the communication and computing resources used in these systems are
safe and reliable. Several research directions have been identified to address the safety
and performance of communication and computing resources in AIoT.
One research direction focuses on improving the reliability of signal characteristics.
In [69], the authors propose a method to reflect the processing methods of the signal
characteristics and internal relations to improve their reliability. They suggest that
this technology can be further improved using big data and 6G technology.
Another research direction aims to improve the performance of CPUs and SoC
implementations for IoT devices. In [58], the authors propose to improve RISC-V CPU
performance for IoT devices, which can enhance their computing power and overall
performance.
Extended architectures are also being explored to enable edge clusters and dynamic
load repartition [17]. This research direction aims to create more flexible and efficient
computing systems that can adapt to changing demands and improve performance.
Furthermore, in [42], the authors aim to investigate the influence of the parameters
associated with the agent called Child and the real consequences for performance. This
research direction aims to optimize the performance of AI agents in IoT systems.
Finally, it is essential to consider the constraints that AI faces when it is imple-
mented in the context of smart buildings. This research direction is discussed in
[17], where the authors aim to model systems with extended architectures that allow
collaboration between many edge devices and dynamic loading with safety.
In general, these research directions focus on improving the safety and performance
of communication and computing resources in AIoT, which is crucial to ensure that
these systems are reliable, efficient, and safe to use in various domains.

4.2 summary
Of the approximately 365 scientific articles performed by the authors, the following
gaps were detected:

36
 • How to ensure that the data read by sensors are without errors/accurate?
• How to update the model’s parameters/attributes automatically?
• How can the integration of sustainable computing with intelligent power and
power systems modeling impact efficiency and sustainability on operations?
• How to optimize energy efficiency in industrial IoT applications to promote
sustainable practices and improve operational performance?
Of the detected gaps, five research directions were formulated.
• Edge AI for increasing autonomy of IoT systems: Apply AI techniques
directly to IoT devices to make them more autonomous and capable of mak-
ing local decisions or processing, without excessively dependent on external
resources, can result in several gains, from reduced latency, efficiency in net-
work use, overload reduction in Cloud, increased autonomy, reinforced privacy,
offline applications, resilience under limited connectivity conditions, among other
benefits.
• Green AIoT: Green Aiot represents a holistic approach to integrating tech-
nological advances, such as artificial intelligence and internet of things, with
environmental concern, seeking innovative and sustainable solutions for con-
temporary challenges (energy efficiency, use of renewable energy, green cities,
sustainable agriculture, monitoring air and water quality, recycling of electronic
components).
• Ethics and Human-in-the-loop: The combination of ethics and the active
presence of human beings in the AI life cycle seeks to ensure that these technolo-
gies are developed and used responsibly, respecting fundamental ethical values
and protecting the interests of affected people. Therefore, certain aspects should
be taken into account such as privacy and data protection, equity and impartiality,
human involvement in decision making, among others.
• Security and Privacy in AIoT: Integrated AIOT security and privacy
approach is essential to ensure users’ confidence and systems integrity, espe-
cially when dealing with intelligent devices interconnection and real -time data
analysis. Therefore the site should ensure protection against cyber threats, data
encryption, security updates, authentication and authorization, resilience against
attacks, among other guarantees.
• Safety and Performance of Communication and Computing Resources
in AIoT: Considering the safety and performance of communication and com-
puting resources in AIOT environments is crucial to ensure the effectiveness,
reliability and integrity of these systems in constant evolution. These aspects are
fundamental to providing a positive experience to users and ensuring the proper
functionality of AIOT applications. For this purpose, periodic software updates
and bandwidth management, mainly.

5 Related Works
The focus of this research is unique and has limited relevance to previous studies.
Multiple review articles have been proposed in the past and some of them have been
found and excluded (ExC3) in this study. Major surveys are briefly introduced in the

37
rest of this section. Fredj et al. [8] provides a comprehensive review of current state-
of-the-art methodologies for designing intelligent IoT systems. The authors analyze
various design methodologies and their strengths and weaknesses. Alshamrani [126]
surveys the current research on the use of IoT and artificial intelligence for remote
healthcare monitoring systems. The authors explore various types of healthcare moni-
toring system, their benefits and the challenges associated with their implementation.
Sabireen and Neelanarayanan [127] presents an in-depth review of fog computing, its
architecture, and its integration with IoT. The authors discuss various algorithms used
in fog computing and current research challenges. Lim and Rahmani [128] presents a
survey of current research on semantic IoT load inference attention management for
healthcare and public health collaboration. The authors explore various approaches
for semantic IoT load inference and attention management. Uddin et al. [129] provides
a comprehensive survey of current research on the adoption of blockchain technology
in the IoT. The authors explore various challenges and solutions associated with the
integration of blockchain technology in IoT. Khanh Duy et al. [130] presents a sur-
vey of current research on IoT data analytics with semantic approaches. The authors
explore various approaches for semantic IoT data analytics and their applications.
Bolhasani et al. [131] presents a systematic review of current research in deep learning
applications for IoT in healthcare. The authors explore various applications of deep
learning in healthcare, including disease diagnosis, prediction, and monitoring. Lim
et al. [132] presents a comprehensive review of smart cities with big data. The authors
explore various reference models, challenges, and considerations associated with the
implementation of smart cities. Rahman et al. [63] presents an intelligent waste man-
agement system using deep learning with IoT. The authors explore various approaches
for waste management using deep learning with IoT. Rey et al. [133] presents a fed-
erated learning approach for malware detection in IoT devices. The authors explore
the challenges associated with malware detection in IoT devices and the potential of
federated learning to address these challenges. Driss et al. [14] presents a review of
microservices in IoT security. The authors explore current solutions, research chal-
lenges, and future directions in the use of microservices for IoT security. Rupani and
Doshi [134] provides a review of current research on smart parking using the IoT. The
authors explore various approaches to smart parking using the IoT and their bene-
fits. Shah et al. [135] presents a survey of smart city infrastructure in New York. The
authors explore various components of the smart city infrastructure, their benefits,
and the challenges associated with their implementation.

6 Conclusion
This review has highlighted the various types of AI models and architectures that are
used in IoT research. The major key findings are:
• The use of edge computing and artificial intelligence in devices provides several
advantages such as improved accuracy of data and behaviors, enhanced safety of
IoT devices and quality of service, increased security inside and between nodes,
greater efficiency of the AIoT nodes, and more sustainable and green computing.

38
 • A large number of types of AI models have been used in the reviewed articles.
The first type of AI model used in the reviewed articles is machine learning. It
includes models such as artificial neural networks, decision trees, and support
vector machines. The second type is MAS, which involves multiple autonomous
agents that interact to achieve common goals. The third type is related to com-
puter vision. The fourth type is knowledge modeling, which involves representing
knowledge using ontologies.
• The application domains of edge computing and AI are diverse, ranging from
healthcare and smart cities to manufacturing and transportation, indicating
a wide range of potential applications and benefits for various industries and
sectors.
Between the major finds we cite also the remain challenges finding:
• Coordination between Agents and IoT Devices: Synchronize and coordinate com-
munication between agents and IoT devices efficiently, especially in dynamic and
heterogeneous environments.
• Scale Management: Handle the expansion of the number of agents and IoT
devices, ensuring operational efficiency and performance in large-scale environ-
ments.
• Interoperability: Overcome interoperability issues between different protocols and
standards used by agents and IoT devices, enabling effective integration.
• Adaptive Decision-Making: Develop adaptive decision-making algorithms that
take into account the dynamics of the IoT environment, changes in conditions,
and interactions between agents.
• Resource Consumption: Efficiently manage resource consumption, such as energy
and bandwidth, to ensure sustainability and operational efficiency.
These findings suggest that the integration of Edge Computing and AI has significant
potential for improving the efficiency, reliability, and sustainability of various sys-
tems and devices and can lead to new and innovative solutions in various application
domains. The quality of the reviewed articles is generally high, with a solid rationale
for the work, an adequate description of the context, a clear statement of the findings
and a discussion of the limitations of the study.
Among the four main research directions, we would like to highlight those related
to Green AIoT and Edge AI for increasing the autonomy of IoT systems. In the
case of Green AIoT, research focuses on making homes and cities more sustainable
by implementing modernization methods such as AI and IoT. Some of the proposed
solutions include smart batteries for homes, cognitive radio sensor networks, machine
learning initiatives for predicting energy consumption, and utilizing wireless commu-
nication standards for better water quality monitoring. On the other hand, edge AI
to increase the autonomy of IoT systems aims to enable intelligent and autonomous
IoT deployments. The research direction focuses on how to make smart homes more
autonomous by deploying machine learning applications to edge servers and employ-
ing service accuracy as a metric in traffic offloading. Other proposed solutions include
incorporating methodologies for converting PET scan images to voxels, improving clas-
sification accuracies on real fish farms, and implementing new alternative microservices
for large-scale IoT applications. Overall, these research directions aim to create more

39
sustainable and efficient homes and cities and increase the autonomy of IoT systems
by incorporating AI and machine learning algorithms to optimize system performance.
Acknowledgments. The first author is partially funded by the Mobility Grant
185AG0B210029 of the Ph.D. program of the French Embassy in Angola. This research
is supported and funded by the Research Cluster R19098 of Zayed University (Dubai,
United Arab Emirates) awarded to the third author.

Declarations
• Funding:
– Pedro Hilaro is partly funded by the Mobility Grant 185AG0B210029 of the
PhD program of the French Embassy in Angola.
– Fatma Outay is supported by the Research Cluster R19098 of Zayed
University (Dubai, United Arab Emirates).
• Conflict of interest/Competing interests: No conflict of interest
• Ethics approval: Not applicable
• Consent to participate: Not applicable
• Consent for publication: All authors and funding schemes have accepted to
publish this article.
• Availability of data and materials: Not applicable
• Code availability: Not applicable
• Authors’ contributions:
– Pedro Hilario is the main author. He has contributed to the definition of the
methodology, the reading of the articles, and the building of internal analysis
documents. He has also contributed to the synthesis of the study in this
article.
– Zeina Elrawashdeh contributed to the definition of the methodology, the read-
ing of the articles, and the construction of internal analysis documents. She
has also contributed to the synthesis of the study in this article.
– Fatma Outay has contributed to the definition of the methodology, the syn-
thesis of the study in this article, and the elaboration of the text related to
the research directions.
– Stéphane Galland has contributed to the definition of the methodology, to
the reading of the articles, the synthesis of the study in this article, and the
elaboration of the text related to the research directions.
– Igor Tchappi contributed to the writing and validation of the text of this
article.

References
[1] Wooldridge, M.: An Introduction to Multiagent Systems, 2nd edn. John wiley
& sons, USA (2009)

[2] Uhrmacher, A.M., Weyns, D.: Multi-Agent Systems: Simulation and Applica-
tions. CRC Press, Taylor & Francis, London, UK (2009)

40
 [3] Simon, H.A.: The Science of Artificial, 3rd edn. MIT Press, Cambridge,
Massachusetts (1996)

[4] Li, W., Logenthiran, T., Phan, V.-T., Woo, W.L.: Intelligent housing develop-
ment building management system (hdbms) for optimized electricity bills. In:
17th IEEE International Conference on Environment and Electrical Engineering
and 2017 1st IEEE Industrial and Commercial Power Systems Europe, EEEIC /
I and CPS Europe, p. 12 (2017). https://doi.org/10.1109/EEEIC.2017.7977410

[5] Tchappi Haman, I., Mualla, Y., Galland, S., Bottaro, A., Kamla, V.C., Kam-
gang, J.-C.: Multilevel and holonic model for dynamic holarchy management:
Application to Large-Scale road traffic. Engineering Applications of Artifi-
cial Intelligence 109, 104622 (2022) https://doi.org/10.1016/j.engappai.2021.
104622

[6] Basso, G., Gaud, N., Gechter, F., Hilaire, V., Lauri, F.: A framework for quali-
fying and evaluating smart grids approaches: Focus on multi-agent technologies.
Smart Grid and Renewable Energy 4(4) (2013) https://doi.org/10.4236/shri.
2013.44040

[7] Feng, J., Hu, X.: An IoT-based hierarchical control method for greenhouse
seedling production. In: Procedia Computer Science; Proceedings of the 25th
International Conference KES2021: Knowledge-Based and Intelligent Informa-
tion & Engineering Systems, vol. 192, pp. 1954–1963. Elsevier, Szczecin, Poland
(2021). https://doi.org/10.1016/j.procs.2021.08.201

[8] Fredj, N., Kacem, Y.H., Khriji, S., Kanoun, O., Abid, M.: A review on intelligent
iot systems design methodologies. In: Measurement: Sensors, vol. 18. Elsevier
Ltd, Yokoama, Japan (2021). https://doi.org/10.1016/j.measen.2021.100347

[9] De Souza, B.P., Motta, R.C., Travassos, G.H.: Towards the description and
representation of smartness in IoT scenarios specification. In: Proceedings of
the XXXIII Brazilian Symposium on Software Engineering. SBES ’19, pp. 511–
516. Association for Computing Machinery, New York, NY, USA (2019). https:
//doi.org/10.1145/3350768.3351797

[10] Galster, M., Weyns, D., Tofan, D., Michalik, B., Avgeriou, P.: Variability in soft-
ware systems - a systematic literature review. IEEE Trans. Softw. Eng. 40(15),
282–306 (2014) https://doi.org/10.1109/TSI.2013.56

[11] Bassam, N.A., Hussain, S.A., Qaraghuli, A.A., Khan, J., Sumesh, E.P., Lavanya,
V.: Iot based wearable device to monitor the signs of quarantined remote patients
of covid-19. Informatics in Medicine Unlocked 24 (2021) https://doi.org/10.
1016/j.imu.2021.100588

[12] Mourtzis, D., Angelopoulos, J., Panopoulos, N., Kardamakis, D.: A smart IoT
platform for oncology patient diagnosis based on ai: Towards the human digital

41
 twin. In: Procedia CIRP, 54th CIRP CMS 2021 - Towards Digitalized Manufac-
turing 4.0, vol. 104. Athen, Greece, pp. 1686–1691 (2021). https://doi.org/10.
1016/j.procir.2021.11.284

[13] Thangaraj, M., Ponmalar, P.P., Sujatha, G., Anuradha, S.: Agent based seman-
tic internet of things (IoT) in smart health care. In: Proceedings of the The
11th International Knowledge Management in Organizations Conference on
The Changing Face of Knowledge Management Impacting Society, vol. Part
F130520. Association for Computing Machinery, New York, NY, USA (2016).
https://doi.org/10.1145/2925995.2926023

[14] Driss, M., Hasan, D., Boulila, W., Ahmad, J.: Microservices in IoT security:
Current solutions, research challenges, and future directions. In: Procedia Com-
puter Science, vol. 192, pp. 2385–2395 (2021). https://doi.org/10.1016/j.procs.
2021.09.007

[15] Rholam, O., Tabaa, M., Monteiro, F., Dandache, A.: Smart device for multi-band
industrial iot communications. In: Procedia Computer Science; The 16th Inter-
national Conference on Mobile Systems and Pervasive Computing (MobiSPC
2019),The 14th International Conference on Future Networks and Communi-
cations (FNC-2019),The 9th International Conference on Sustainable Energy
Information Technology, vol. 155, pp. 660–665. Elsevier B.V., Halifax, Canada
(2019). https://doi.org/10.1016/j.procs.2019.08.094

[16] Prutyanov, V., Melentev, N., Lopatkin, D., Menshchikov, A., Somov, A.: Devel-
oping IoT devices empowered by artificial intelligence: Experimental study. In:
2019 Global IoT Summit (GIoTS), pp. 1–6. IEEE Communications Society,
Aarhus, Denmark (2019). https://doi.org/10.1109/GIOTS.2019.8766355

[17] Debauche, O., Mahmoudi, S., Mahmoudi, S.A., Manneback, P., Lebeau, F.: A
new edge architecture for AI-IoT services deployment. In: Procedia Computer
Science; The 17th International Conference on Mobile Systems and Pervasive
Computing (MobiSPC),The 15th International Conference on Future Networks
and Communications (FNC),The 10th International Conference on Sustainable
Energy Information Technology, vol. 175, pp. 10–19. Elsevier B.V., Leuven,
Belgium (2020). https://doi.org/10.1016/j.procs.2020.07.006

[18] Budgen, D., Brereton, P.: Performing systematic literature reviews in software
engineering. In: Proceedings of the 28th International Conference on Software
Engineering, pp. 1051–1052 (2006). ACM

[19] Kitchenham, B.A., Brereton, P., Turner, M., Niazi, M.K., Linkman, S., Pre-
torius, R., Budgen, D.: Refining the systematic literature review process—two
participant-observer case studies. Empirical Software Engineering 15(6), 618–
653 (2010) https://doi.org/10.1007/s10664-010-9134-8

[20] Tchappi Haman, I., Galland, S., Kamla, V.C., Kamgang, J.-C., Mualla, Y.,

42
 Najjar, A., Hilaire, V.: A critical review of holonic technology in traffic and trans-
portation fields. Engineering Applications of Artificial Intelligence 1-54(90),
1–54 (2020) https://doi.org/10.1016/j.engappai.2020.103503

[21] Mualla, Y., Najjar, A., Daoud, A., Galland, S., Nicolle, C., Yasar, A.-U.-H.,
Shakshuki, E.: Agent-based simulation of unmanned aerial vehicles in civil-
ian applications: A systematic literature review and research directions. Future
Generation Computer Systems 100, 344–364 (2019) https://doi.org/10.1016/j.
future.2019.04.051

[22] Brereton, P., Kitchenham, B.A., Budgen, D., Turner, M., Khalil, M.: Lessons
from applying the systematic literature review process within the software
engineering domain. Journal of Systems and Software 80(16), 571–583 (2007)
https://doi.org/10.1016/j.jss.2006.05.009

[23] Gusenbauer, M., Haddaway, N.R.: Which academic search systems are suitable
for systematic reviews or meta-analyses? evaluating retrieval qualities of google
scholar, PubMed, and 26 other resources. Research Synthesis Methods 11(2),
181–217 (2020) https://doi.org/10.1002/jrsm.1378

[24] Calvaresi, D., Cesarini, D., Sernani, P., Marinoni, M., Dragoni, A.F., Sturm, A.:
Exploring the ambient assisted living domain: a systematic review. Journal of
Ambient Intelligence and Humanized Computing 35(17), 4–30 (2017)

[25] Jain, R., Suman, U.: A systematic literature review on global software develop-
ment life cycle. ACM Digital library (portal.acm.org/dl.cfm) 14(13), 2–13 (2015)
https://doi.org/10.1145/2735399.2735408

[26] Kitchenham, B., Brereton, O.P., Budgen, D., Turner, M., Bailey, J., Linkman,
S.: Systematic literature reviews in software engineering – a systematic literature
review. ACM Digital library (portal.acm.org/dl.cfm) 14(114), 2–13 (2008)

[27] Kitchenham, B.: Guidelines for performing systematic literature reviews in soft-
ware engineering. ResearchGate 66(18), 2–20 (2007) https://doi.org/10.1016/j.
infsof.2008.09.009

[28] Calvaresi, D., Marinoni, M., Sturm, A., Schumacher, M., Buttazzo, G.: The
challenge of real-time multi-agent systems for enabling IoT and CPS. In: Pro-
ceedings of the International Conference on Web Intelligence, pp. 356–364.
Association for Computing Machinery, New York, NY, USA (2017). https:
//doi.org/10.1145/3106426.3106518

[29] Markets & Markets: Internet of things (IoT) market by software solution
(real-time streaming analytics, security solution, data management, remote mon-
itoring, and network bandwidth management), platform, service, application
domain, and region - global forecast to 2022. Technical report, Markets&Markets
(2017). https://www.marketsandmarkets.com/Market-Reports/iot-market-573.

43
 html

[30] McKinsey Global Institute: The internet of things: Mapping the value beyond
the hype. Technical report, McKinsey Global Institute (2019)

[31] Ibrahim, A.S., Al-Mahdi, H., Nassar, H.: Characterization of task response time
in a fog-enabled IoT network using queueing models with general service times.
Journal of King Saud University - Computer and Information Sciences 34(9),
7089–7100 (2022) https://doi.org/10.1016/j.jksuci.2021.09.008

[32] Kousiouris, G., Tsarsitalidis, S., Psomakelis, E., Koloniaris, S., Bardaki, C.,
Tserpes, K., Nikolaidou, M., Anagnostopoulos, D.: A microservice-based frame-
work for integrating IoT management platforms, semantic and AI services for
supply chain management. ICT Express 5(2), 141–145 (2019) https://doi.org/
10.1016/j.icte.2019.04.002

[33] Choudhary, S., Kesswani, N.: Analysis of KDD-Cup’99, NSL-KDD and UNSW-
NB15 datasets using deep learning in IoT. In: Procedia Computer Science;
Proceedings of the International Conference on Computational Intelligence and
Data Science, vol. 167, pp. 1561–1573 (2020). https://doi.org/10.1016/j.procs.
2020.03.367

[34] Verdouw, C., Sundmaeker, H., Tekinerdogan, B., Conzon, D., Montanaro, T.:
Architecture framework of IoT-based food and farm systems: A multiple case
study. Computers and Electronics in Agriculture 165, 104939 (2019) https://
doi.org/10.1016/j.compag.2019.104939

[35] Galaz, V., Centeno, M.A., Callahan, P.W., Causevic, A., Patterson, T., Brass, I.,
Baum, S., Farber, D., Fischer, J., Garcia, D., McPhearson, T., Jimenez, D., King,
B., Larcey, P., Levy, K.: Artificial intelligence, systemic risks, and sustainability.
Technology in Society 67, 101741 (2021) https://doi.org/10.1016/j.techsoc.2021.
101741

[36] Le, K.-H., Le-Minh, K.-H., Thai, H.-T.: BrainyEdge: An AI-enabled framework
for IoT edge computing. ICT Express 9(2), 211–221 (2023) https://doi.org/10.
1016/j.icte.2021.12.007

[37] Klingensmith, N., Kim, Y., Banerjee, S.: A hypervisor-based privacy agent
for mobile and IoT systems. In: Proceedings of the 20th International Work-
shop on Mobile Computing Systems and Applications. HotMobile’19, pp. 21–26.
Association for Computing Machinery, New York, NY, USA (2019). https:
//doi.org/10.1145/3301293.3302356

[38] Rahman, M.S., Khalil, I., Yi, X., Atiquzzaman, M., Bertino, E.: A lossless data-
hiding based IoT data authenticity model in edge-AI for connected living. ACM
Trans. Internet Technol. 22(3) (2021) https://doi.org/10.1145/3453171

44
[39] Agossou, B.E., Toshiro, T.: IoT & AI based system for fish farming: Case study
of benin. In: Proceedings of the Conference on Information Technology for Social
Good. GoodIT ’21, pp. 259–264. Association for Computing Machinery, New
York, NY, USA (2021). https://doi.org/10.1145/3462203.3475873

[40] Babu, N.T.R., Stewart, C.: Energy, latency and staleness tradeoffs in AI-driven
IoT. In: Proceedings of the 4th ACM/IEEE Symposium on Edge Computing.
SEC ’19, pp. 425–430. Association for Computing Machinery, New York, NY,
USA (2019). https://doi.org/10.1145/3318216.3363381

[41] Ramesh, M.V.: Integration of participatory approaches, systems, and solutions

using IoT and AI for designing smart community: Case studies from india. In:
Proceedings of the 1st ACM International Workshop on Technology Enablers
and Innovative Applications for Smart Cities and Communities. TESCA’19, pp.
4–5. Association for Computing Machinery, New York, NY, USA (2019). https:
//doi.org/10.1145/3364544.3371501

[42] De Souza, J.T., Campos, G.A.L., Rocha, C., Werbet, E., Costa, L.F.d., Melo,
R.T., Alves, L.V.: An agent program in an IoT system to recommend activities
to minimize childhood obesity problems. In: Proceedings of the 35th Annual
ACM Symposium on Applied Computing. SAC ’20, pp. 654–661. Association for
Computing Machinery, New York, NY, USA (2020). https://doi.org/10.1145/
3341105.3373927

[43] Chehri, A., Zimmermann, A., Schmidt, R., Masuda, Y.: Theory and practice of
implementing a successful enterprise IoT strategy in the industry 4.0 era. In:
Procedia Computer Science; Proceedings of the 25th International Conference
KES2021: Knowledge-Based and Intelligent Information & Engineering Systems,
vol. 192, pp. 4609–4618 (2021). https://doi.org/10.1016/j.procs.2021.09.239

[44] Sun, W., Liu, J., Yue, Y.: AI-enhanced offloading in edge computing: When
machine learning meets industrial IoT. IEEE Network 33(5), 68–74 (2019) https:
//doi.org/10.1109/MNET.001.1800510

[45] Kamalakkannan, S., Kulatunga, A.K., Bandara, L.: The conceptual framework
of IoT based decision support system for life cycle management. In: Proce-
dia Manufacturing; Proceedings of the 17th Global Conference on Sustainable
Manufacturing: Sustainable Manufacturing - Hand in Hand to Sustainability on
Globe, vol. 43, pp. 423–430 (2020). https://doi.org/10.1016/j.promfg.2020.02.
192

[46] Latif, S.A., Wen, F.B.X., Iwendi, C., Wang, L.-l.F., Mohsin, S.M., Han, Z.,
Band, S.S.: AI-empowered, blockchain and SDN integrated security architecture
for IoT network of cyber physical systems. Computer Communications 181,
274–283 (2022) https://doi.org/10.1016/j.comcom.2021.09.029

[47] Chen, S., Gong, P., Wang, B., Anpalagan, A., Guizani, M., Yang, C.: EDGE AI

45
 for heterogeneous and massive IoT networks. In: 2019 IEEE 19th International
Conference on Communication Technology (ICCT), pp. 350–355 (2019). https:
//doi.org/10.1109/ICCT46805.2019.8947193

[48] Ragavi, B., Pavithra, L., Sandhiyadevi, P., Mohanapriya, G.K., Harikirubha, S.:
Smart agriculture with AI sensor by using agrobot. In: 2020 Fourth International
Conference on Computing Methodologies and Communication (ICCMC), pp.
1–4 (2020). https://doi.org/10.1109/ICCMC48092.2020.ICCMC-00078

[49] Vij, A., Vijendra, S., Jain, A., Bajaj, S., Bassi, A., Sharma, A.: IoT and
machine learning approaches for automation of farm irrigation system. In:
Procedia Computer Science; Proceedings of the International Conference on
Computational Intelligence and Data Science, vol. 167, pp. 1250–1257 (2020).
https://doi.org/10.1016/j.procs.2020.03.440

[50] Chehri, A., Chaibi, H., Saadane, R., Hakem, N., Wahbi, M.: A framework of
optimizing the deployment of IoT for precision agriculture industry. In: Procedia
Computer Science; Proceedings of the 24th International Conference KES2020:
Knowledge-Based and Intelligent Information & Engineering Systems, vol. 176,
pp. 2414–2422 (2020). https://doi.org/10.1016/j.procs.2020.09.312

[51] Debauche, O., Mahmoudi, S., Elmoulat, M., Mahmoudi, S.A., Manneback, P.,
Lebeau, F.: Edge AI-IoT pivot irrigation, plant diseases, and pests identification.
In: Procedia Computer Science; Proceedings of the 11th International Confer-
ence on Emerging Ubiquitous Systems and Pervasive Networks (EUSPN 2020)
/ The 10th International Conference on Current and Future Trends of Informa-
tion and Communication Technologies in Healthcare (ICTH 2020) / Affiliated
Workshops, vol. 177, pp. 40–48 (2020). https://doi.org/10.1016/j.procs.2020.10.
009

[52] Chen, C.-J., Huang, Y.-Y., Li, Y.-S., Chang, C.-Y., Huang, Y.-M.: An AIoT
based smart agricultural system for pests detection. IEEE Access 8, 180750–
180761 (2020) https://doi.org/10.1109/ACCESS.2020.3024891

[53] Kaushik, I., Prakash, N.: Applicability of IoT for smart agriculture: Challenges
& future research direction. In: 2021 IEEE World AI IoT Congress (AIIoT), pp.
0462–0467 (2021). https://doi.org/10.1109/AIIoT52608.2021.9454209

[54] Torai, S., Chiyoda, S., Ohara, K.: Application of AI technology to smart agricul-
ture: Detection of plant diseases. In: 2020 59th Annual Conference of the Society
of Instrument and Control Engineers of Japan (SICE), pp. 1514–1519 (2020).
https://doi.org/10.23919/SICE48898.2020.9240353

[55] Gia, T.N., Qingqing, L., Queralta, J.P., Zou, Z., Tenhunen, H., Westerlund,
T.: Edge AI in smart farming IoT: CNNs at the edge and fog computing
with LoRa. In: 2019 IEEE AFRICON, pp. 1–6 (2019). https://doi.org/10.1109/
AFRICON46755.2019.9134049

46
[56] Chukkapalli, S.S.L., Mittal, S., Gupta, M., Abdelsalam, M., Joshi, A., Sandhu,
R., Joshi, K.: Ontologies and artificial intelligence systems for the cooperative
smart farming ecosystem. IEEE Access 8, 164045–164064 (2020) https://doi.
org/10.1109/ACCESS.2020.3022763

[57] Bruneo, D., Distefano, S., Giacobbe, M., Longo Minnolo, A., Longo, F., Merlino,
G., Mulfari, D., Panarello, A., Patanè, G., Puliafito, A., Puliafito, C., Tapas, N.:
An IoT service ecosystem for smart cities: The #smartme project. Internet of
Things 5, 12–33 (2019) https://doi.org/10.1016/j.iot.2018.11.004

[58] Liu, Z., Jiang, J., Lei, G., Chen, K., Qin, B., Zhao, X.: A heterogeneous processor
design for CNN-based AI applications on IoT devices. In: Procedia Computer
Science; Proceedings of the 2019 International Conference on Identification,
Information and Knowledge in the Internet of Things, vol. 174, pp. 2–8 (2020).
https://doi.org/10.1016/j.procs.2020.06.048

[59] Saqib, M., Jasra, B., Moon, A.H.: A lightweight three factor authentication
framework for IoT based critical applications. Journal of King Saud University
- Computer and Information Sciences 34(9), 6925–6937 (2022) https://doi.org/
10.1016/j.jksuci.2021.07.023

[60] Hazarika, A., Poddar, S., Nasralla, M.M., Rahaman, H.: Area and energy
efficient shift and accumulator unit for object detection in IoT applications.
Alexandria Engineering Journal 61(1), 795–809 (2022) https://doi.org/10.1016/
j.aej.2021.04.099

[61] Bordin, C., Håkansson, A., Mishra, S.: Smart energy and power systems mod-
elling: an IoT and cyber-physical systems perspective, in the context of energy
informatics. In: Procedia Computer Science; Proceedings of the 24th Interna-
tional Conference KES2020: Knowledge-Based and Intelligent Information &
Engineering Systems, vol. 176, pp. 2254–2263 (2020). https://doi.org/10.1016/
j.procs.2020.09.275

[62] Jo, W., Shin, Y., Kim, H., Yoo, D., Kim, D., Kang, C., Jin, J., Oh, J., Na, B.,
Shon, T.: Digital forensic practices and methodologies for AI speaker ecosystems.
Digital Investigation 29, 80–93 (2019) https://doi.org/10.1016/j.diin.2019.04.
013

[63] Rahman, W., Islam, R., Hasan, A., Bithi, N.I., Hasan, M., Rahman, M.M.:
Intelligent waste management system using deep learning with iot. Journal of
King Saud University - Computer and Information Sciences 34(5), 2072–2087
(2022) https://doi.org/10.1016/j.jksuci.2020.08.016

[64] Santa, J., Bernal-Escobedo, L., Sanchez-Iborra, R.: On-board unit to connect
personal mobility vehicles to the IoT. In: Procedia Computer Science; Proceed-
ings of the 17th International Conference on Mobile Systems and Pervasive
Computing (MobiSPC),The 15th International Conference on Future Networks

47
 and Communications (FNC),The 10th International Conference on Sustainable
Energy Information Technology, vol. 175, pp. 173–180 (2020). https://doi.org/
10.1016/j.procs.2020.07.027

[65] Zekić-Sušac, M., Mitrović, S., Has, A.: Machine learning based system for man-
aging energy efficiency of public sector as an approach towards smart cities.
International Journal of Information Management 58, 102074 (2021) https:
//doi.org/10.1016/j.ijinfomgt.2020.102074

[66] Manman, L., Goswami, P., Mukherjee, P., Mukherjee, A., Yang, L., Ghosh,
U., Menon, V.G., Qi, Y., Nkenyereye, L.: Distributed artificial intelligence
empowered sustainable cognitive radio sensor networks: A smart city on-demand
perspective. Sustainable Cities and Society 75, 103265 (2021) https://doi.org/
10.1016/j.scs.2021.103265

[67] Soh, Z.H.C., Kamarulazizi, K., Daud, K., Hamzah, I.H., Saad, Z., Abdul-
lah, S.A.C.: Abandoned baggage detection & alert system via AI and IoT.
In: Proceedings of the 2020 12th International Conference on Computer and
Automation Engineering. ICCAE 2020, pp. 205–209. Association for Comput-
ing Machinery, New York, NY, USA (2020). https://doi.org/10.1145/3384613.
3384614

[68] Anh Khoa, T., Phuc, C.H., Lam, P.D., Nhu, L.M.B., Trong, N.M., Phuong,
N.T.H., Dung, N.V., Tan-Y, N., Nguyen, H.N., Duc, D.N.M.: Waste management
system using IoT-based machine learning in university. Wireless Communica-
tions and Mobile Computing 6138637 (2020) https://doi.org/10.1155/2020/
6138637

[69] Lv, Z., Qiao, L., Kumar Singh, A., Wang, Q.: AI-empowered IoT security for
smart cities. ACM Trans. Internet Technol. 21(4) (2021) https://doi.org/10.
1145/3406115

[70] Alladi, T., Kohli, V., Chamola, V., Yu, F.R., Guizani, M.: Artificial intelligence
(AI)-empowered intrusion detection architecture for the internet of vehicles.
IEEE Wireless Communications 28(3), 144–149 (2021) https://doi.org/10.1109/
MWC.001.2000428

[71] Soomro, S., Miraz, M.H., Prasanth, A., Abdullah, M.: Artificial intelligence
enabled IoT: Traffic congestion reduction in smart cities. In: Smart Cities
Symposium 2018, pp. 1–6 (2018). https://doi.org/10.1049/cp.2018.1381

[72] Karapetyan, A., Chau, S.C.-K., Elbassioni, K., Azman, S.K., Khonji, M.: Multi-
sensor adaptive control system for IoT-empowered smart lighting with oblivious
mobile sensors. ACM Trans. Sen. Netw. 16(1) (2019) https://doi.org/10.1145/
3369392

[73] Li, Z., Wang, Y., Wang, W., Greuter, S., Mueller, F.F.: Empowering a creative

48
 city: Engage citizens in creating street art through human-AI collaboration. In:
Extended Abstracts of the 2020 CHI Conference on Human Factors in Comput-
ing Systems. CHI EA ’20, pp. 1–8. Association for Computing Machinery, New
York, NY, USA (2020). https://doi.org/10.1145/3334480.3382976

[74] Backhouse, J.: A taxonomy of measures for smart cities. In: Proceedings of the
13th International Conference on Theory and Practice of Electronic Governance.
ICEGOV ’20, pp. 609–619. Association for Computing Machinery, New York,
NY, USA (2020). https://doi.org/10.1145/3428502.3428593

[75] Roppestad, R., Fyhn, P.-G., Colomo-Palacios, R.: A test bed for smart energy
education in the field of computer engineering. In: Proceedings of the Second
International Conference on Technological Ecosystems for Enhancing Multicul-
turality. TEEM ’14, pp. 101–105. Association for Computing Machinery, New
York, NY, USA (2014). https://doi.org/10.1145/2669711.2669886

[76] Firouzi, R., Rahmani, R., Kanter, T.: An autonomic IoT gateway for smart home
using fuzzy logic reasoner. In: Procedia Computer Science; Proceedings of the
11th International Conference on Emerging Ubiquitous Systems and Pervasive
Networks (EUSPN 2020) / The 10th International Conference on Current and
Future Trends of Information and Communication Technologies in Healthcare
(ICTH 2020) / Affiliated Workshops, vol. 177, pp. 102–111 (2020). https://doi.
org/10.1016/j.procs.2020.10.017

[77] Ashraf, S.: A proactive role of IoT devices in building smart cities. Internet of
Things and Cyber-Physical Systems 1, 8–13 (2021) https://doi.org/10.1016/j.
iotcps.2021.08.001

[78] Zia, M.: B-drive: A blockchain based distributed iot network for smart urban
transportation. Blockchain: Research and Applications 2(4), 100033 (2021)
https://doi.org/10.1016/j.bcra.2021.100033

[79] Lim, C., Cho, G.-H., Kim, J.: Understanding the linkages of smart-city tech-
nologies and applications: Key lessons from a text mining approach and a call
for future research. Technological Forecasting and Social Change 170, 120893
(2021) https://doi.org/10.1016/j.techfore.2021.120893

[80] Nespoli, P., Dı́az-López, D., Gómez Mármol, F.: Cyberprotection in IoT envi-
ronments: A dynamic rule-based solution to defend smart devices. Journal of
Information Security and Applications 60, 102878 (2021) https://doi.org/10.
1016/j.jisa.2021.102878

[81] Dubey, S., Singh, P., Yadav, P., Singh, K.K.: Household waste management
system using IoT and machine learning. In: Procedia Computer Science; Pro-
ceedings of the International Conference on Computational Intelligence and Data
Science, vol. 167, pp. 1950–1959 (2020). https://doi.org/10.1016/j.procs.2020.
03.222

49
[82] Koumetio Tekouabou, S.C., Abdellaoui Alaoui, E.A., Cherif, W., Silkan, H.:
Improving parking availability prediction in smart cities with IoT and ensemble-
based model. Journal of King Saud University - Computer and Information
Sciences 34(3), 687–697 (2022) https://doi.org/10.1016/j.jksuci.2020.01.008

[83] Oberascher, M., Kinzel, C., Kastlunger, U., Kleidorfer, M., Zingerle, C., Rauch,
W., Sitzenfrei, R.: Integrated urban water management with micro storages
developed as an iot-based solution - the smart rain barrel. Environmental Mod-
elling & Software 139, 105028 (2021) https://doi.org/10.1016/j.envsoft.2021.
105028

[84] Chopvitayakun, S., Jantamala, S.: IOT smart home for elderly and unattended
residence. In: Proceedings of the 11th International Conference on Education
Technology and Computers. ICETC ’19, pp. 322–326. Association for Comput-
ing Machinery, New York, NY, USA (2020). https://doi.org/10.1145/3369255.
3369284

[85] Lakshmikantha, V., Hiriyannagowda, A., Manjunath, A., Patted, A., Basavaiah,
J., Anthony, A.A.: IoT based smart water quality monitoring system. In: Global
Transitions Proceedings of the International Conference on Computing System
and Its Applications (ICCSA- 2021), vol. 2, pp. 181–186 (2021). https://doi.org/
10.1016/j.gltp.2021.08.062

[86] Bagdadee, A.H., Hoque, M.Z., Zhang, L.: IoT based wireless sensor network for
power quality control in smart grid. In: Procedia Computer Science; Proceedings
of the International Conference on Computational Intelligence and Data Science,
vol. 167, pp. 1148–1160 (2020). https://doi.org/10.1016/j.procs.2020.03.417

[87] Pasika, S., Gandla, S.T.: Smart water quality monitoring system with cost-
effective using IoT. Heliyon 6(7) (2020) https://doi.org/10.1016/j.heliyon.2020.
e04096

[88] Wang, J.X., Liu, Y., Lei, Z.-b., Wu, K.-h., Zhao, X.-y., Feng, C., Liu, H.-w.,
Shuai, X.-h., Tang, Z.-m., Wu, L.-y., Long, S.-y., Wu, J.-r.: Smart water lora IoT
system. In: Proceedings of the 2018 International Conference on Communication
Engineering and Technology. ICCET ’18, pp. 48–51. Association for Comput-
ing Machinery, New York, NY, USA (2018). https://doi.org/10.1145/3194244.
3194260

[89] Vlachostergiou, A., Stratogiannis, G., Caridakis, G., Siolas, G., Mylonas, P.:
Smart home context awareness based on smart and innovative cities. In: Pro-
ceedings of the 16th International Conference on Engineering Applications of
Neural Networks (INNS). EANN ’15. Association for Computing Machinery,
New York, NY, USA (2015). https://doi.org/10.1145/2797143.2797150

[90] Crabbe, C.: Smart parking for future smart cities using fog computing

50
 paradigms. In: Proceedings of the 7th International Conference on ICT for Sus-
tainability. ICT4S2020, pp. 193–195. Association for Computing Machinery, New
York, NY, USA (2020). https://doi.org/10.1145/3401335.3401358

[91] Bekara, C.: Security issues and challenges for the IoT-based smart grid. In: Pro-
cedia Computer Science; Proceedings of the 9th International Conference on
Future Networks and Communications (FNC’14)/The 11th International Con-
ference on Mobile Systems and Pervasive Computing (MobiSPC’14)/Affiliated
Workshops, vol. 34, pp. 532–537 (2014). https://doi.org/10.1016/j.procs.2014.
07.064

[92] Tapas, N., Vyas, O.P.: IoT deployment for smarter cities with special reference
to mobility. In: Proceedings of the Second International Conference on Internet
of Things, Data and Cloud Computing. ICC ’17. Association for Comput-
ing Machinery, New York, NY, USA (2017). https://doi.org/10.1145/3018896.
3036390

[93] Valero, C.I., Gil, A.M.M., Gonzalez-Usach, R., Julian, M., Fico, G., Arredondo,
M.T., Stavropoulos, T.G., Strantsalis, D., Voulgaridis, A., Roca, F., Jara, A.J.,
Serrano, M., Zappa, A., Khan, Y., Guillen, S., Sala, P., Belsa, A., Votis, K.,
Palau, C.E.: AIoTES: Setting the principles for semantic interoperable and
modern IoT-enabled reference architecture for active and healthy ageing ecosys-
tems. Computer Communications 177, 96–111 (2021) https://doi.org/10.1016/
j.comcom.2021.06.010

[94] Bourgais, M., Giustozzi, F., Vercouter, L.: Detecting situations with stream
reasoning on health data obtained with IoT. In: Procedia Computer Science;
Proceedings of the 25th International Conference KES2021: Knowledge-Based
and Intelligent Information & Engineering Systems, vol. 192, pp. 507–516 (2021).
https://doi.org/10.1016/j.procs.2021.08.052

[95] Al Bassam, N., Hussain, S.A., Al Qaraghuli, A., Khan, J., Sumesh, E.P.,
Lavanya, V.: IoT based wearable device to monitor the signs of quarantined
remote patients of COVID-19. Informatics in Medicine Unlocked 24, 100588
(2021) https://doi.org/10.1016/j.imu.2021.100588

[96] Bouchareb, A., Boulaalam, A., Bellamine, I.: IoT and AI based intelligent system
to fight against COVID-19. In: Proceedings of the 4th International Confer-
ence on Networking, Information Systems & Security. NISS2021. Association for
Computing Machinery, New York, NY, USA (2021). https://doi.org/10.1145/
3454127.3457618

[97] Talukder, A., Haas, R.: AIoT: AI meets IoT and web in smart healthcare. In:
Companion Publication of the 13th ACM Web Science Conference 2021. Web-
Sci’21 Companion, pp. 92–98. Association for Computing Machinery, New York,
NY, USA (2021). https://doi.org/10.1145/3462741.3466650

51
 [98] Vangipuram, S.k., Appusamy, R.: Machine learning framework for COVID-
19 diagnosis. In: International Conference on Data Science, E-Learning and
Information Systems 2021. DATA’21, pp. 18–25. Association for Comput-
ing Machinery, New York, NY, USA (2021). https://doi.org/10.1145/3460620.
3460624

[99] Srivastava, V., Ruchilekha: Diagnosing COVID-19 using AI based medical image
analysis. In: 5th Joint International Conference on Data Science & Management
of Data (9th ACM IKDD CODS and 27th COMAD). CODS-COMAD 2022, pp.
204–212. Association for Computing Machinery, New York, NY, USA (2022).
https://doi.org/10.1145/3493700.3493730

[100] Nakamura, K., Manzoni, P., Redondi, A., Longo, E., Zennaro, M., Cano, J.-C.,
Calafate, C.T.: A LoRa-based protocol for connecting IoT edge computing nodes
to provide small-data-based services. Digital Communications and Networks
8(3), 257–266 (2022) https://doi.org/10.1016/j.dcan.2021.08.007

[101] Fang, Z., Fu, H., Gu, T., Qian, Z., Jaeger, T., Hu, P., Mohapatra, P.: A model
checking-based security analysis framework for IoT systems. High-Confidence
Computing 1(1), 100004 (2021) https://doi.org/10.1016/j.hcc.2021.100004

[102] Haddad, H., Bouyahia, Z., Chaudhry, S.A.: A multiagent geosimulation and
IoT-based framework for safety monitoring in complex dynamic spatial environ-
ments. In: Procedia Computer Science; Proceedings of the 10th International
Conference on Ambient Systems, Networks and Technologies (ANT 2019) / The
2nd International Conference on Emerging Data and Industry 4.0 (EDI40 2019)
/ Affiliated Workshops, vol. 151, pp. 527–534 (2019). https://doi.org/10.1016/
j.procs.2019.04.071

[103] Bideh, P.N., Sönnerup, J., Hell, M.: Energy consumption for securing lightweight
IoT protocols. In: Proceedings of the 10th International Conference on the Inter-
net of Things. IoT ’20. Association for Computing Machinery, New York, NY,
USA (2020). https://doi.org/10.1145/3410992.3411008

[104] Wazid, M., Bera, B., Mitra, A., Das, A.K., Ali, R.: Private blockchain-envisioned
security framework for AI-enabled IoT-based drone-aided healthcare services. In:
Proceedings of the 2nd ACM MobiCom Workshop on Drone Assisted Wireless
Communications for 5G and Beyond. DroneCom ’20, pp. 37–42. Association for
Computing Machinery, New York, NY, USA (2020). https://doi.org/10.1145/
3414045.3415941

[105] Okamoto, K., Nebashi, R., Banno, N., Bai, X., Numata, H., Iguchi, N., Miya-
mura, M., Hada, H., Funahashi, K., Sugibayashi, T., Sakamoto, T., Tada, M.:
ON-state retention of atom switch eNVM for IoT/AI inference solution. In:
2020 IEEE International Reliability Physics Symposium (IRPS), pp. 1–4 (2020).
https://doi.org/10.1109/IRPS45951.2020.9128967

52
[106] Indrawati, Anantha, I.M., Amani, H.: Identification of smart technology indi-
cators for measuring smart city: An indonesian perspective. In: Proceedings of
the 2017 International Conference on Telecommunications and Communication
Engineering. ICTCE ’17, pp. 75–80. Association for Computing Machinery, New
York, NY, USA (2017). https://doi.org/10.1145/3145777.3145782

[107] Zhou, H., Zhang, H., Zhou, Y., Wang, X., Li, W.: Botzone: An online multi-
agent competitive platform for AI education. In: Proceedings of the 23rd Annual
ACM Conference on Innovation and Technology in Computer Science Education.
ITiCSE 2018, pp. 33–38. Association for Computing Machinery, New York, NY,
USA (2018). https://doi.org/10.1145/3197091.3197099

[108] Shaw, N.P., Stöckel, A., Orr, R.W., Lidbetter, T.F., Cohen, R.: Towards prov-
ably moral AI agents in bottom-up learning frameworks. In: Proceedings of
the 2018 AAAI/ACM Conference on AI, Ethics, and Society. AIES ’18, pp.
271–277. Association for Computing Machinery, New York, NY, USA (2018).
https://doi.org/10.1145/3278721.3278728

[109] Srivastava, B., Rossi, F.: Towards composable bias rating of ai services. In:
Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society.
AIES ’18, pp. 284–289. Association for Computing Machinery, New York, NY,
USA (2018). https://doi.org/10.1145/3278721.3278744

[110] Wang, D., Weisz, J.D., Muller, M., Ram, P., Geyer, W., Dugan, C., Tausczik,
Y., Samulowitz, H., Gray, A.: Human-AI collaboration in data science: Explor-
ing data scientists’ perceptions of automated AI. Proc. ACM Hum.-Comput.
Interact. 3(CSCW) (2019) https://doi.org/10.1145/3359313

[111] Hadfield-Menell, D., Hadfield, G.K.: Incomplete contracting and AI alignment.

In: Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society.
AIES ’19, pp. 417–422. Association for Computing Machinery, New York, NY,
USA (2019). https://doi.org/10.1145/3306618.3314250

[112] McKee, H.A., Porter, J.E.: Ethics for AI writing: The importance of rhetorical
context. In: Proceedings of the AAAI/ACM Conference on AI, Ethics, and Soci-
ety. AIES ’20, pp. 110–116. Association for Computing Machinery, New York,
NY, USA (2020). https://doi.org/10.1145/3375627.3375811

[113] Ortiz, G., Boubeta-Puig, J., Criado, J., Corral-Plaza, D., Garcia-de-Prado, A.,
Medina-Bulo, I., Iribarne, L.: A microservice architecture for real-time IoT
data processing: A reusable web of things approach for smart ports. Computer
Standards & Interfaces 81, 103604 (2022) https://doi.org/10.1016/j.csi.2021.
103604

[114] Mohamudally, N., Peermamode-Mohaboob, M.: Building an anomaly detection

engine (ADE) for IoT smart applications. In: Procedia Computer Science; Pro-
ceedings of the 15th International Conference on Mobile Systems and Pervasive

53
 Computing (MobiSPC 2018) / The 13th International Conference on Future
Networks and Communications (FNC-2018) / Affiliated Workshops, vol. 134,
pp. 10–17 (2018). https://doi.org/10.1016/j.procs.2018.07.138

[115] Osamy, W.M., Khedr, A., Salim, A., Al Ali, A., El-Sawy, A.: A review on recent
studies utilizing artificial intelligence methods for solving routing challenges in
wireless sensor networks. PeerJ Computer Science 8(e1089) (2022) https://doi.
org/10.7717/peerj-cs.1089

[116] Cao, L., Cai, Y., Yue, Y.: Swarm intelligence-based performance optimization
for mobile wireless sensor networks: Survey, challenges, and future directions.
IEEE Access 7, 161524–161553 (2019) https://doi.org/10.1109/ACCESS.2019.
2951370

[117] Kaveh, M., Mesgari, M.S.: Application of meta-heuristic algorithms for train-
ing neural networks and deep learning architectures: A comprehensive review.
Neural Processing Letters (2022) https://doi.org/10.1007/s11063-022-11055-6

[118] Buyya, R., Broberg, J., Goscinski, A.: Cloud Computing: Concepts, Technology
& Architecture, 1st edition edn. Pearson Education, London, UK (2013). https:
//doi.org/10.1007/978-981-19-3026-3

[119] Gubbi, J., Buyya, R., Marusic, S., Palaniswami, M.: Internet of things (IoT):
A vision, architectural elements, and future directions. Future Generation Com-
puter Systems 29(7), 1645–1660 (2013) https://doi.org/10.26483/IJARCS.V7I7.
6082

[120] Bonomi, F., Milito, R., Zhu, J., Addepalli, S.: Fog computing and its role in
the internet of things. In: Proceedings of the First Edition of the MCC Work-
shop on Mobile Cloud Computing, pp. 13–16 (2012). https://doi.org/10.4018/
978-1-5225-7149-0.ch003 . ACM

[121] Armbrust, M., Fox, A., Griffith, R., Joseph, A.D., Katz, R.H., Konwinski, A.,
Lee, G., Patterson, D.A., Rabkin, A., Stoica, I., Zaharia, M.: A view of cloud
computing. Communications of the ACM 53(4), 50–58 (2013) https://doi.org/
10.2991/ijndc.2013.1.1.2

[122] Satyanarayanan, M., Bahl, P., Caceres, R., Davies, N.: The case for VM-based
cloudlets in mobile computing. IEEE Pervasive Computing 8(4), 14–23 (2009)
https://doi.org/10.1109/MPRV.2009.82

[123] Alam, M., Rufino, J., Ferreira, J., Ahmed, S.H., Shah, N., Chen, Y.: Orchestra-
tion of microservices for iot using docker and edge computing. IEEE Communi-
cations Magazine 56(9), 118–123 (2018) https://doi.org/10.1109/MCOM.2018.
1701233

54
[124] Wang, K., Zhao, Y., Gangadhari, R.K., Li, Z.: Analyzing the adoption chal-
lenges of the internet of things (IoT) and artificial intelligence (AI) for smart
cities in china. Sustainability 13(19), 10983 (2021) https://doi.org/10.3390/
su131910983

[125] D’Aquin, M., Troullinou, P., O’Connor, N.E., Cullen, A., Faller, G., Holden,
L.: Towards an ”ethics by design” methodology for ai research projects. In:
Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society.
AIES ’18, pp. 54–59. Association for Computing Machinery, New York, NY,
USA (2018). https://doi.org/10.1145/3278721.3278765

[126] Alshamrani, M.: Iot and artificial intelligence implementations for remote health-
care monitoring systems: A survey. Journal of King Saud University - Computer
and Information Sciences 34(8, Part A), 4687–4701 (2022) https://doi.org/10.
1016/j.jksuci.2021.06.005

[127] Sabireen, H., Neelanarayanan, V.: A review on fog computing: Architecture, fog
with IoT, algorithms and research challenges. ICT Express 7(2), 162–176 (2021)
https://doi.org/10.1016/j.icte.2021.05.004

[128] Lim, S., Rahmani, R.: Toward semantic iot load inference attention manage-
ment for facilitating healthcare and public health collaboration: A survey. In:
Procedia Computer Science; Proceedings of the 11th International Conference
on Emerging Ubiquitous Systems and Pervasive Networks (EUSPN 2020) / The
10th International Conference on Current and Future Trends of Information and
Communication Technologies in Healthcare (ICTH 2020) / Affiliated Workshops,
vol. 177, pp. 371–378 (2020). https://doi.org/10.1016/j.procs.2020.10.050

[129] Uddin, M.A., Stranieri, A., Gondal, I., Balasubramanian, V.: A survey on the
adoption of blockchain in iot: challenges and solutions. Blockchain: Research and
Applications 2(2), 100006 (2021) https://doi.org/10.1016/j.bcra.2021.100006

[130] Khanh Duy, T., Küng, J., Huu Hanh, H.: Survey on iot data analytics with
semantic approaches. In: The 23rd International Conference on Information
Integration and Web Intelligence. iiWAS2021, pp. 199–204. Association for
Computing Machinery, New York, NY, USA (2022). https://doi.org/10.1145/
3487664.3487785

[131] Bolhasani, H., Mohseni, M., Rahmani, A.M.: Deep learning applications for
IoT in health care: A systematic review. Informatics in Medicine Unlocked 23,
100550 (2021) https://doi.org/10.1016/j.imu.2021.100550

[132] Lim, C., Kim, K.-J., Maglio, P.P.: Smart cities with big data: Reference models,
challenges, and considerations. Cities 82, 86–99 (2018) https://doi.org/10.1016/
j.cities.2018.04.011

[133] Rey, V., Sánchez Sánchez, P.M., Huertas Celdrán, A., Bovet, G.: Federated

55
 learning for malware detection in iot devices. Computer Networks 204, 108693
(2022) https://doi.org/10.1016/j.comnet.2021.108693

[134] Rupani, S., Doshi, N.: A review of smart parking using internet of things
(IoT). In: Procedia Computer Science; Proceedings of the the 10th International
Conference on Emerging Ubiquitous Systems and Pervasive Networks (EUSPN-
2019) / The 9th International Conference on Current and Future Trends of
Information and Communication Technologies in Healthcare (ICTH-2019) /
Affiliated Workshops, vol. 160, pp. 706–711 (2019). https://doi.org/10.1016/j.
procs.2019.11.023

[135] Shah, J., Kothari, J., Doshi, N.: A survey of smart city infrastructure via
case study on new york. In: Procedia Computer Science; Proceedings of the
10th International Conference on Emerging Ubiquitous Systems and Perva-
sive Networks (EUSPN-2019) / The 9th International Conference on Current
and Future Trends of Information and Communication Technologies in Health-
care (ICTH-2019) / Affiliated Workshops, vol. 160, pp. 702–705 (2019). https:
//doi.org/10.1016/j.procs.2019.11.024

56