This article has been accepted for publication in IEEE Open Journal of the Communications Society.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

Received XX Month, XXXX; revised XX Month, XXXX; accepted XX Month, XXXX; Date of publication XX Month, XXXX; date of
current version XX Month, XXXX.
Digital Object Identifier 10.1109/OJCOMS.2022.1234567

XAI for Industry 5.0 - Concepts,

Opportunities, Challenges and Future
Directions
THIPPA REDDY GADEKALLU∗ , PRAVEEN KUMAR REDDY MADDIKUNTA† ,
PRABADEVI BOOPATHY∗ , NATARAJAN DEEPA∗ , RAJESWARI CHENGODEN‡ , NANCY
VICTOR∗ , WEI WANG§ , WEIZHENG WANG¶ , YAODONG ZHU∥ , and KAPAL DEV∗∗
1
The College of Mathematics and Computer Science, Zhejiang A&F University, Hangzhou 311300, China. Division of Research and
Development, Lovely Professional University, Phagwara, India Center of Research Impact and Outcome, Chitkara University, Punjab, India.
(e-mail: [email protected]).
2
School of Computer Science Engineering and Information Systems, Vellore Insititute of Technology, Vellore, India.
(e-mail: [email protected], [email protected], [email protected], [email protected])
3
Department of Computer Science, Pondicherry University, Karaikal Campus, Karaikal - 609605, Puducherry Union Territory, India.
(email: [email protected]).
4
Guangdong-Hong Kong-Macao Joint Laboratory for Emotion Intelligence and Pervasive Computing, Artificial Intelligence Research
Institute, Shenzhen MSU-BIT University, Shenzhen 518172, China. (e-mail: [email protected]).
5
Department of Computer Science, City University of Hong Kong, Hong Kong SAR, China (e-mail: [email protected]).
6
School of Information Science and Engineering, Jiaxing University, Jiaxing, China (email: [email protected]).
7
Department of Computer Science and CONNECT Centre, Munster Technological University, Bishopstown, Cork, T12 P928, Ireland.
(e-mail: [email protected]).
CORRESPONDING AUTHOR: Yaodong Zhu (e-mail: [email protected]).

ABSTRACT Industry 5.0 has become a reality now and it is a paradigm that integrates contemporary
innovations and concepts. Artificial Intelligence (AI) is a key component and asset of the industrial
transformation which allows intelligent devices to perform functionalities such as self-examination,
assessment, and evaluation autonomously. AI-based methodologies using ML and deep learning assist
manufacturers and industrialists in forecasting service requirements and minimizing downtime. Recent
research has discovered a remarkable change in the processes, systems, applications, and products in
industries. Also, there is a significant challenge with the explainability of the decisions provided by
the models using deep learning algorithms and their inadequate ability to be coupled with each other.
Therefore, Explainable artificial intelligence (XAI) is required without compromising the efficiency of the
models developed using deep learning algorithms. XAI investigates and develops algorithms, techniques,
and models that produce human-comprehensible explanations of AI-based systems and can increase
transparency and performance. The explainability nature of XAI will help humans understand the model
and the reason behind the predictions, thus improving the model’s transparency and the reliability of the
outcomes. Furthermore, an Industry 5.0-enabled environment has a variety of data from varied sources,
and this multi-source information must be fused to derive meaningful and optimal decisions. Therefore, all
AI-integrated applications must derive actionable insights through information fusion. Hence, the adoption
of XAI methodologies in Industry 5.0 can help humans make trustworthy decisions for critical applications
requiring information fusion. In this paper, we present a state-of-the-art survey on adopting XAI in Industry
5.0. We discuss the adoption of XAI in various applications such as smart factories, smart Healthcare,
E-Governance, smart transportation, Education 5.0, Agriculture 5.0, and Energy 5.0. Finally, some research
issues and future directions of integrating XAI with Industry 5.0 are also discussed and highlighted to
promote more study in the potential field.

INDEX TERMS XAI, AI, Industry 5.0, Smart Factories, Smart Healthcare, E-Governance, Smart Trans-
portation, Education 5.0, Agriculture 5.0 and Energy 5.0.

VOLUME , 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

I. Introduction environment. Manufacturers can increase their operational

The origin of Industry 5.0 (I5.0) may be traced back to the intelligence by adopting information fusion [7]. For example,
manufacturing sector’s movement from the era of Industrial data from production floor sensors can be amalgamated with
Revolution to the age of digital transformation and beyond. insights from quality control, supply chain logistics, and
Each successive stage highlights the industrial revolution’s customer feedback to improve production processes, predict
growth in the production process, which has altered the maintenance needs, and customize products to satisfy spe-
approach of working in the industry [1]. The Industrial Rev- cific customer needs. This improves efficiency and produc-
olution has significantly transformed economies that were tivity and enables greater customization and creativity, which
initially dependent on agriculture to the present generation meets I5.0’s fundamental principles. Furthermore, informa-
of large-scale industries [2]. With consistent technologi- tion fusion enables humans and machines to communicate
cal advancement, new production methods and techniques seamlessly, allowing real-time adjustments and collaborative
have come into existence, reducing cumbersome human problem-solving. This dynamic interaction promotes a more
efforts and marking significant enhancement in production, robust and flexible production system that can respond to
economy and overall societal standards. Although industrial dynamic consumer preferences and technological advance-
manufacturing techniques have evolved throughout the years, ments [8].
the fundamental purpose of industry has remained the same The fundamental computing technologies of Industry 4.0
to provide goods that improve the quality of life. Radical include AI, ML, Big Data, Cloud Computing, and cyber-
concepts have occasionally surfaced to alter the direction of security, whereas Automation and Robotics, IoT, CPS, and
industry throughout the history of manufacturing [3]. Since AM constitute the physical part [9]. Together, these tech-
the inception of Industry 1.0 in the eighteenth century involv- nologies realize the advantages of Industry 4.0 systems to
ing steam power, there have been giant leaps of progression enable agile, adaptable, and on-demand manufacturing, a
to the present era with the application of information and crucial component of smart factories or smart manufacturing
communication technologies in various industrial processes. frameworks. Although implementing these technologies can
Network connectivity has enabled production systems in- help businesses gain a competitive edge and increase op-
volving complex computing technologies to be expanded erational efficiency, there is widespread concern that doing
to the development of digital twins in this generation. so could result in the loss of jobs for low-skilled workers
This has ensured communication across diversified facilities due to the high level of automation, which could lead to an
and helped acquire data, leading to production automation. imbalance in the economy and greater inequality in society
The networking of all systems has created ”cyber-physical [10]. Also, complete automation can generate false positive
production systems” and ”smart factories,” where people, or false negative output, resulting in a waste of resources in
components, and production systems interact over a network the production process. Thus, despite the ability to digitally
and produce almost autonomously. Industry 4.0 has revealed connect machines, enabling a seamless flow of data and
its ability to bring forth some fantastic advancements in optimization, I5.0 is considered to reinstate the human touch
manufacturing environments when these aforementioned en- in the process to ensure sustainable manufacturing [11],
ablers have come together [4]. For example, it has enabled [12]. Therefore, the need for collaboration between humans,
machines to foresee problems and autonomously initiate robots, and digital technologies has been established.
repair procedures or self-organized logistics that respond to Various studies have emphasised the inception, perception,
unforeseen changes in production. The production environ- applications, and challenges of I5.0. For example, the survey
ment thus has become more digitalized, which has further in [13] debated the various aspects of I4.0 and I5.0 and
opened up more flexible ways to deliver the appropriate discussed various opportunities and challenges about their
information to the right person at the right time. Digital co-existence. The survey in [14] acts as a whistle-blower
devices inside production units and in the field have ensured about the global transformation involving I5.0. Humans and
that maintenance professionals are equipped with relevant their well-being are placed at the core of the manufacturing
service history and documentation at the right time when systems and the three most important characteristics of I5.0
they need to use them. This has also reduced wastage of namely human-centricity, sustainability and resiliency are
time, providing scope for the maintenance professionals to discussed. The study presented a tri-dimensional perspective
invest their efforts exclusively on problem resolution [5]. of I5.0 implementation: technical, reality, and application
Information fusion is an essential element in achieving dimensions. In [15], the predominant enabling technologies
the goals of I5.0. Information fusion involves integrating and applications of I5.0 are highlighted, and further, a
data from various sources, including IoT, sensors, third-party survey-based tutorial on the applications of I5.0 is presented.
databases, and human inputs, to produce a more accurate, The study by [16] performed an exhaustive survey on pub-
complete, and responsive view of the manufacturing pro- lished articles having the keyword ”I5.0” and text mining
cess [6]. This integrated method for data processing aims was implemented to extract key terms and perform frequency
to improve decision-making, increase production ability, analysis in this regard. The key enablers of I5.0 were
and create a more responsive and adaptable manufacturing thus identified, and various application areas, namely sup-

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

2 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

ply chain evaluation, enterprise innovation and digitization, tions on service sectors like smart factories, healthcare,
smart manufacturing, and various other areas, were explored governance, transportation, education, agriculture, and
further. The study revealed the need for human-machine energy. The survey is the first of its kind dealing
connectivity and co-existence as a significant theme for I5.0- with versatile applications in the depth of the problem
based sustainable development. The various studies on I5.0 domain
have highlighted its potential and applications involving key • Technical and non-technical requirements of the above-
enabling technologies in association with human interven- said applications are discussed with practical insight.
tion, and associated challenges have also surfaced that are This would foster the researchers to work on the key
relevant to the transparency and traceability of the generated areas of various applications using XAI.
outcomes. In I5.0 handling the data from different sources • The proposed survey discussed various challenges re-
necessitates using more transparent and understandable AI garding the application of XAI, since it is a white-
systems. XAI is essential for explaining AI-made decisions, box model that may affect the data’s privacy, integrity
supporting trustworthiness, and guaranteeing compliance. and security aspects and the point of contact with
XAI elucidates the clear reasoning behind decisions made, smart industry. This is addressed with the enhanced
assisting users in comprehending AI-powered recommenda- elucidation with the scope analysis of futuristic research
tions that are vital for making safe, economic, and oper- • The proposed survey discusses various use cases of XAI
ational decisions. Additionally, XAI addresses inconsisten- in smart industries, with detailed discussions regarding
cies in data and facilitates collaboration between humans the societal contribution of XAI and I5.0.
and machines by ensuring optimal transparency and ease • The proposed survey analyses the possibility of Man-
of use in AI operations. In addition, XAI guarantees that Machine parallel living with the support of the XAI as
data fusion adheres to regulatory and ethical guidelines by the tool for the interaction and communication between
offering complete audit trails, thereby promoting compliance the entities, with the enhanced capacity of elucidating
and accountability in complex industrial situations. Table. 1 to the end user about the perception of AI engines.
depicts the Summary of Literature Survey.
The present survey thus emphasizes providing an ex- B. Organization of the Survey
haustive survey on the potential of XAI in I5.0 and rel- The proposed survey is organized with the following sec-
evant applications. XAI, often called Interpretable AI or tions. Section.II deals with the evolution of the XAI with
Explainable Machine Learning (XML), is a type of AI the description of the types and processes associated with
in which humans can comprehend the justification for the it. Section.III discusses about various applications of the
decisions or predictions produced by the AI. It is a col- XAI in real-life demands such as Smart factory, Health
lection of procedures and techniques that make it possible care, Governance, Transportation, Education and Energy
for human users to understand and believe the output and with the required discussion of use-cases and applications.
results produced by machine learning (ML) algorithms. It Section.IV provides various challenges in integrating I5.0
thus helps to describe an AI model and understand the with XAI by discussing possible solutions and futuristic
expected impact and potential biases. Also, it ensures the recommendations. Section.V concludes with the motivation
characterization of model accuracy and transparency of the for the futuristic research with the Integration of I5.0 and
outcomes involved in the AI-based decision-making process. XAI.
Furthermore, I5.0, connecting numerous heterogeneous de-
vices and applications coordinated with human input, may
II. Background
accumulate vast volumes of data through information fusion
A. History of Industry X.0
that could be used to build AI models. The interpretation
of the model, its decisions, and various data biases must be The industrial revolution disseminated the work from human
understood for ease of use, safety, and a better understanding hands to machines, eventually leading to the mass produc-
of these models’ outcomes and data usage. Hence, XAI tion of goods. In the late 17th century, people discovered
would suffice this need for I5.0’s seamless adoption and the existence of abundant natural fossil fuels and innova-
successful collaboration of humans with machines. The use tive machines that drastically transformed human society’s
of XAI is exceptionally relevant in the context of I5.0-based livelihood. In 1712, Thomas Newcomen invented the first
applications, as building trust and confidence is crucial while atmospheric engine that harnessed the steam that is produced
putting AI-based models into production. by boiling water to convert it to mechanical energy [17].
Initially, this design was used to pump out the water from
the mines which significantly reduced human efforts. Later,
A. Contributions of the Proposed Survey
in 1776, James Watt a Scottish inventor, improved the design
. The significant contributions of the survey are listed below:
of the steam engine [18] which paved a futuristic way for the
• The proposed survey presents the research by provid- concept of thermodynamics. This invention also has become
ing various possible applications in which XAI can the fundamental design that significantly broadens its usage
be deployed with in-depth analysis of XAI applica- and revolutionized it across other fields like manufacturing,

VOLUME , 3

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

TABLE 1. Summary of Literature Survey

Ref.No. Contribution Limitation

[13] -The characteristics of Industry 4.0 and the transition to I5.0 are discussed. -The primary emphasis was the comparative analysis of Industry 4.0 and I5.0.
-The issues pertaining to the coexistence of Industry 4.0 and I5.0 are addressed -I5.0, and its associated challenges were briefly discussed without much emphasis on
plausible means of improvement.
[14] The human-centricity, sustainability, and resiliency aspects of I5.0 is discussed. Societal issues in I5.0 implementation and challenges pertaining to the science and
-A three-dimensional architecture of implementing I5.0 is presented considering the technology aspects are discussed without highlighting potential enablers to improve the
technical dimension, reality dimension and application dimension same
[11] -The complementation of I5.0 and co-evolution of I5.0 and Society 5.0 are explored. The shared challenges associated with Human Cyber-Physical Systems, Greentelligent
-The Goal, Value, Organization and Technological dimension pertaining to comparison Manufacturing (GIM), Human-Robot Collaboration are discussed but the technologies
of Industry 4.0 and Society 5.0 are highlighted establishing the dependency of one with to combat the challenges are not explored
the other.
[15] -I5.0 concepts, the enabling technologies, and applications are discussed in detail. The generic challenges are enlisted and discussed superficially lagging in-depth analysis
-Research challenges and open issues in I5.0 implementations are identified and and exploration on potential remedial enablers for the same.
highlighted
[16] -Text mining tools are used to filter all publications where the “Industry 5.0” terminology The challenges pertaining to the various applications are not explored especially in the
is used. realm of AI, IoT and relevant enablers
-The application of I5.0 in supply chain evaluation and optimization, enterprise
innovation and digitization, smart manufacturing is discussed
This Paper A comprehensive survey on Industry 5.0, its evolution stages and the relevant enabling The proposed CAI and I5.0 integration’s implementation constraints and benefits should
technologies are discussed. I5.0 applications are primarily dependent on AI based be evaluated in further study.
technologies and the associated accuracy, transparency and interpretability issues are
highlighted. The use of XAI to overcome such challenges is given prime importance in
this paper, and its contributions to improving I5.0 applications are presented.

steam turbines in power generation plants, locomotives, and used microprocessor-based hardware technology. Industry
steamships. 3.0 was a huge leap that allowed the use of Programmable
The first industrial revolution was named Industry 1.0 Logic Controllers (PLC), robots, etc, to automate industry
(I1.0), which started in Britain in 1784 and spread across processes. It was during this period of transformation that
the globe in the early 18th century. I1.0 utilized fossil more robots were employed, mainly in industries, to operate
fuels like coal, oil, and natural gas. When these fuels are the tasks that were performed by humans. However, there
burnt, they release enormous amounts of energy converted was still human interference behind it.
to mechanical energy. I1.0 was driven majorly by water- The human participation in the automation processes in
powered machines, which led to more incredible innovations, I3.0 was totally replaced with the up gradation of I3.0
such as steamships, that changed the perception of ordinary to Industry 4.0 (I4.0) which was referred to as an era of
people about time and space. Following I1.0 almost a century Cyber-Physical Systems (CPS). In I3.0 the machines were
later, in the early 19th century, the world was stepping into only automatized but the data generated from them were
the second industrial revolution, which was named Industry to be carried forward by the humans for further decision
2.0 (I2.0). This is also referred to as the “Technological makings and necessary actions were to be taken based on
Revolution” as many new technological innovations such the decisions. CPS is basically the integration of hardware
as the internal combustion engine, effective methods of and software components that are connected together for
communication such as the telegraph and the telephone, seamless communication to perform the defined tasks. CPS
railroad network, gas and water supply, chemical synthesis, is defined as the addition of physical systems like sensors
and mass production in manufacturing and consumer goods, and actuators, cyber systems like communication, computa-
the first automobile and the first aeroplane were all invented. tion, and storage, and protocols, on the internet technology.
The I2.0 was between 1870 and 1914, which enabled the Examples of CPS include intelligent robots that make a plan
transportation of goods in large volumes from one place to of action by themselves, intelligent buildings that control
another and paved the way for globalization. everything from lighting, and energy usage to user-specific
After the I2.0, it was an era of computers. The first and functions, smart autonomous vehicles that drive themselves,
second-generation computers that used vacuum tubes and and planes that fly themselves in controlled airspace [19].
transistors, respectively, came into existence during 1940- I4.0 brought advances in intelligent machines, however,
1960, and modern computers took their shape with the I4.0 has failed in cases such as the COVID-19, pandemic
advent of time. The third industrial revolution, Industry 3.0 period when the entire world was at a standstill. Though
(I3.0), started slowly, and remarkable inventions happened industries have started adopting I4.0, by which the entire
after 1969 and were referred to as the “digital revolution”. process is fully automated still the machines did not operate
The third-generation computers that use integrated chip tech- during the pandemic period. The greatest understanding that
nology (ICT) were invented in the 1960s and leveraged fully is derived from the COVID-19 crisis is that machines de-
automated machines to aid human operators or entirely re- pend on human operators, human programmers, and human
place humans. Later the fourth generation (1970) computers maintenance, and not everything could be automated. Human

4 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

cognitive power and analytical capabilities are irreplaceable of machinery, which become an extension of the worker‘s
by machines. Similarly, the conventional methods adopted body. The cobots can be reprogrammed with a tablet allow-
in the field of manufacturing in the era of I4.0 is very ing them to handle different tasks depending on the changes
straightforward. The robots involved in the production phase in the workflow. Such agility has enabled Mercedes to offer
are “dumb” as they would not consider the real users’ personalized designs according to the user’s choices. In this
choice but keep producing the parts in the pipeline and use case, the intelligence of the human and the ability of the
later would assemble it as a final product. This assembly robots are harnessed for improving the flexibility in design,
system could not deliver the customization people wanted. personalization concerning the customer‘s choice, decision-
The I4.0 framework mainly focussed on automation and making in critical situations with human intelligence, and
thereby mass production was possible but customization speeding and scaling up the process with the help of the
was not possible. Later, the same business required the robots [1]. Similarly in healthcare, I5.0 can be used to
collaborative intelligence of both man and machine to enable develop new medical technologies and devices that allow
customization. Therefore, I4.0 evolved into a new concept the medical team to provide accurate and personalized care
called I5.0, focusing on the interaction between man and for individuals. For instance, essential human body vitals are
machine. collected and monitored remotely using wearable devices by
I5.0 refers to human working with robots and smart medical practitioners. This allows the doctors or the nurses
devices. It adds the human touch to I4.0. During I4.0 the to intervene if the patient‘s data condition is alarming.
intention was to minimize human involvement and emphasis
process automation. I5.0 is a new model that believes real B. Key enabling technologies
progress lies in the collaboration between humans and ma- The technology enablers of I5.0 include:
chines. I5.0 does not replace I4.0, rather I5.0 is a compliment
of I4.0. I5.0 framework utilises the features I4.0 and along 1) Human-machine-interaction technologies: I5.0 fos-
with that adds the human operator to the framework. Hence ters a collaborative relationship between humans and
interaction between human and machines are inevitable in machines, leveraging their complementary strengths
I5.0 and focusses on personalization. I5.0 has the vision for increased effectiveness. This is evident in the med-
of an innovative, resilient, socio-centered, and competitive ical field, where I5.0 facilitates the development of in-
industry, which respects planetary boundaries and minimizes clusive tools catering to individual needs. Furthermore,
its negative environmental impact. technologies like AR/VR provide immersive training
I5.0 is the enhanced version of I4.0 that promotes re- experiences, ensuring continued learning even during
silience, is human-centric and is sustainable. The human- pandemics [21]. Additionally, cobots augment human
centric approach promotes the worker’s fundamental rights, capabilities in completing complex tasks, exemplified
the well-being of the worker, and a safe and inclusive by personalized medical equipment that seamlessly
environment. Due to the automation in I4.0, society might adapts to a surgeon’s hand.
feel their jobs are taken away. As a result, industrial workers 2) Bio-inspired technologies and smart materials: Bio-
need to keep upskilling and re-skilling themselves for work- inspired technologies that draw design inspiration from
life balance [20]. I5.0 helps to improve the quality of life nature offer diverse solutions across various fields,
of the individual and society as a whole. Another essential including robotics, control systems, data analysis, and
aspect of I5.0 is the intelligent use of natural resources, optimization. Bio-inspired robotics like gecko-grippers
which are reusable, re-purposed, and re-cycled, thereby pro- and virus-inspired nanosystems for drug delivery [22]
ducing sustainable products reducing waste, and improving demonstrate the adaptability of this approach. Notably,
the utilization of the available resources. The resilience of self-healing technologies inspired by skin regeneration
the current industry infrastructure is not as expected and has hold the potential to minimize waste, extend product
been proven unsuccessful during the recent pandemic period. lifespans, and enhance safety when interacting with hu-
The proposal and adoption of new industry policies should mans [23]. Meanwhile, self-healing materials inspired
enhance the resilience of the industrial system during global by spider silk promise longer-lasting, safer products for
emergencies or uncertainties in the production or supply- applications like prosthetics and wind turbine blades
chain process. [24]. Furthermore, bio-inspired materials such as bio-
The man-machine symbiosis makes remarkable imprints adhesives and biologically made materials could be
in the fields of healthcare, supply chain management, man- used for construction industries, 3D printing, and prod-
ufacturing, automobiles and etc. In manufacturing, for in- uct development, promoting strengthened production.
stance, Mercedes-Benz, to stand apart in the automobile Also, bio-inspired AI algorithms, devised based on
business, invested in flexible manufacturing to offer person- natural processes like Ant colony and swarm, could
alization. AI-enabled collaborative robots (cobots) replace help develop more robust systems.
the robots in the plant. The workers in the manufacturing 3) Real-time based digital twins and simulation to
plant guide the cobot’s arms to pick up and place heavy parts model entire systems: I5.0 is shaped by the digi-
tal twin in a number of ways, including predictive

VOLUME , 5

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

Industry 0
XXX

Industry 2.0
Electricity &
Communication

1870-1914
Telephone, Telegraph,
Industry 1.0
Rise of Aeroplane, Mass Industry 4.0
Manufacturing Production, Steel,
1784-1822

Automobile Internet
Utilized Hydro/nuclear
Weaving Loom
Industry 3.0 Smart Building, Smart

2011
Difference Engine
Vehicles, Smart City,
Automation
Utilized Fossil Fuels Smart Factories
Utilized Sensors

1969
Evolution of Computers Industry 5.0

Utilized ICTs, Electronics Cobots

Personalized
Manufacturing,

2021
Automobile, Healthcare
Cognition of human &
endurance of robots

Industry X.0

FIGURE 1. Evolution of Industry 5.0

maintenance, which involves continuously monitoring [28], mental state monitoring for co-worker well-being,
the systems to foresee failures before they occur, pro- and enhanced rehabilitation through assistive BCI-
cess optimization, which simulates various workable based tools. Numerous sophisticated correlation anal-
methods as a process solution, quality control, which ysis systems can adapt to novel or unforeseen circum-
involves detecting defects early in the manufacturing stances. These encompass meta-learning, counterfac-
process, supply chain management, which involves tual reasoning, and transfer learning. The three primary
better inventory tracking, remote operations, which components of I5.0 are people-oriented, environment-
avoids human intervention in dangerous locations, oriented, and resilience-oriented. To address power
training and simulation, which creates a realistic envi- imbalances and strengthen grid resilience, the author
ronment for improving trainees’ skills, and customer in this study [29] suggested combining deep learning
experience, which involves allowing users to see and and transfer learning models to forecast power output
comment on the product [25]. To achieve human- in photovoltaic power plants.
centricity in smart manufacturing systems and move 5) Technologies for storage, computing, and secure
toward I5.0, the Human Digital Twin (HDT) idea is data transmission: Blockchain, edge, and cloud com-
suggested as a crucial technique in [26]. HDTs are puting technologies are all components of an ecosys-
digital avatars of people to revolutionize the field tem in the modern industrial revolution. Cloud com-
of human-system integration by closely integrating puting offers resources on demand for applications re-
human traits into the architecture and functionality of quiring sophisticated data processing. Critical benefits
the system. including immediate decision-making, privacy in local
4) Artificial Intelligence: In line with the goal of I5.0 data, and greater resilience are made possible in the
of fostering human-machine collaboration, the Brain- edge computing. Blockchain facilitates transparent and
Computer Interface (BCI) acts as a bridge between the safe transactions. The author of this paper [30] offers
two in novel ways [27]. These include direct control an empowered edge computing architecture that uses
over machines for complex tasks, real-time brain signal less energy and allows for real-time decision-making.
capture and analysis through cognitive augmentation The cloud acts as a central repository from which edge

6 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

devices obtain insights. Real-time operations are per- 2) Meaningful: The intended user should be able to
formed by edge devices, which also send critical data understand the explanations provided by the system.
to the cloud for further processing and preservation. 3) Explanation Accuracy: The explanation should accu-
Blockchain transfers data and transactions securely rately reflect the processes of the system and/ or
over the edge and cloud, ensuring transparency and reasons for the predictions/ results provided.
confidence. 4) Knowledge Limits: The XAI-based system should
work only under the conditions for which it was
designed and when it attains the required confidence
C. XAI in its result.
Building accurate AI models is the need of the hour, as
it has applications in all domains ranging from healthcare The significant motivating factors of XAI include trust,
to climate forecasting to autonomous vehicles to particle informativeness, accountability, causability, transferability,
physics. The massive availability of computing resources and fair and ethical decision making and so on. This makes
the increasing data sizes have geared the development of XAI useful for all the stakeholders including the users,
many complex non-linear models. The early phases of data the people who are affected by the decisions of the model
modeling depended upon handcrafted rules and heuristics. and the developers [34].An XAI model could personalize
The linear models and decision trees were then introduced explanations based on the patient’s medical history, explain
to build models pertaining to several applications. However, the impact of lifestyle modifications like diet and exercise
the inception of ensemble and deep learning models have on blood sugar levels or impact of medicines. Adapting to
took over the entire stage by building efficient models. Meta- new information obtained from the patient’s body vitals’, the
learning or the models that has the capability to create other model could provide tailored recommendations for adjusting
models is the recent trend. The development of accurate AI medication or lifestyle habits. Such XAI model could be built
models enabled a paradigm shift across various dimensions on decentralized data without sharing sensitive patient infor-
including expressiveness, versatality, adaptability and effi- mation or could add noise to the data to protect individual
ciency. The sophistication of the AI-enabled systems has privacy while preserving the overall utility of the model.
grown to such an extent where no human intervention is even
required for the designing and deployment of such systems.
However, as the internal complexity of the models increases, 1) Taxonamy of XAI
the more opaque it becomes. The predictions made by the AI XAI can be categorized into various types based on different
models can be non-intuitive thereby making it difficult for dimensions [35]:
the people to understand what the model is actually doing.
This would be a matter of concern when the decisions of the 1) Ante-hoc vs. Post-hoc: This classification is based on
model are used for critical applications such as healthcare, the stage at which expainability is incorporated to the
military and so on. The capabilities of AI models can be system. The ante-hoc model follows a new learning
further enhanced by allowing humans to closely collaborate approach where the explainability is seeded into the
with the machines so that solutions can be devised even model from the initial phase itself; whereas in the post-
for complex problems. Even though ML models are widely hoc approach, a black box phenomenon is followed
used, the ”black box” nature of the model makes the human until the training phase is completed and explainability
doubtful about the trustworthiness of the model when dealing will be incorporated only during the testing time.
with critical applications. Integrating the “explainability” 2) Model-Specific vs. Model-Agnostic: This classification
feature with the AI system will enable humans to understand is based on whether the XAI approach is dependent
the model by explaining why a particular prediction is or independent of the ML model being utilized. It
made, there by increasing the transparency of the model and basically depends on the method’s applicability to
trustworthiness of the results provided [?], [31]. According various models. The model-specific techniques are
to [32] “XAI will create a suite of ML techniques that white box and they function by inspecting the internal
enables human users to understand, appropriately trust, and structure of the model. Model-agnostic techniques in
effectively manage the emerging generation of artificially turn are black box approaches wherein they function
intelligent partners”. The use of white-box models in XAI by inspecting the relationship between input-output
within smart industries causes privacy and security concerns pairs of the models and does not depend on the model
due to their transparency. Table. 2 shows the Privacy and internals.
Security Challenges in White-Box XAI for Smart Industries. 3) Local vs. Global: This classification is based on the
The authors in [33] proposed the four key principles any scope of the explanation provided. Local explainability
XAI-based system should adhere to. This includes: provides explanation about how a particular prediction
is made for a specific input. This is significant in un-
1) Explanation: The system should be able to provide derstanding what all unique features have contributed
valid evidence or reasons for the processes or outputs. to the final decision at a particular instance. Global

VOLUME , 7

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

TABLE 2. Privacy and Security Challenges in White-Box XAI for Smart Industries

Challenge Overview Potential Impact Severity Level Industries Most Affected

Transparency Trade-offs Increased transparency can expose sensitive informa- Loss of competitive advantage, increased risk of High Manufacturing, Aerospace, Elec-
tion about the system, posing IP and misuse risks. reverse-engineering, and potential misuse of propri- tronics, Pharmaceuticals
etary information.
Data Privacy Concerns Detailed explanations may reveal private or sensitive Violation of data privacy regulations, loss of cus- High Healthcare, Finance, Retail,
customer/proprietary data. tomer trust, and potential legal liabilities. Telecommunications
Adversarial Attacks White-box nature makes the system more vulnerable Disruption of critical operations, data breaches, and High Critical Infrastructure, Energy,
to attacks exploiting system details. financial losses. Transportation, Defense
Interpretability Limitations Complexity still makes it challenging to fully explain Difficulty in gaining trust and acceptance from end- Moderate All industries
the system’s reasoning. users, and challenges in decision-making and over-
sight.
Model Complexity Complex algorithms for detailed explanations can Reduced system performance, increased maintenance Moderate Manufacturing, Logistics, Energy,
impact performance and scalability. costs, and limited deployment options. Utilities
Regulatory Compliance Reconciling transparency with protecting sensitive Regulatory fines, legal disputes, and reputational High Healthcare, Finance, Telecommu-
data and trade secrets. damage. nications, Aerospace
User Trust and Acceptance Detailed explanations may not align with user ex- Reduced user adoption, decreased system effective- Moderate All industries
pectations and mental models. ness, and challenges in user training and support.

explainability, on the other hand, aids in deriving the theory. Using this framework, each feature is assigned
features that are significant for the model output as a an importance value for a specific prediction [37].
whole. It also helps in analyzing how a model learns by One of the significant features of SHAP include the
understanding the changes in the extend of a specific determination of a novel class of feature importance
attribute employed in the decision making process of metrics. The authors in [38] proposed an interactive
the model. explainable framework using SHAP. This framework
can be used for providing explanations about power
system emergency control [39], intrusion detection
2) XAI Frameworks [40], network anomaly detection [41] and so on.
XAI frameworks are tools that can be used for generating Apart from the SHAP and LIME frameworks, various other
reports regarding how a particular model functions and algorithms also help in providing explanations about the
provide explanations regarding its functioning. Fig. 2 depicts predictions made by the model [42]. ELI5 is a python
the XAI frameworks. The most popular XAI frameworks, package that helps to examine the underlying ML model
LIME and SHAP are explained below. and explain the predictions provided. Similarly, Facets is
1) LIME: LIME stands for Local Interpretable Model- an algorithm that is model-agnostic and ante-hoc. Here, the
agnostic Explanations. Each part of the name explains explanations can be provided both locally and globally. Fig.
the features of this framework. This uses a model- shows the various XAI algorithms with their specific XAI
agnostic approach and is a post-hoc model. Here, the taxonamy.
scope of explanation is local. The predictions made Even though explainability is not a mandatory feature for
using the LIME framework is individually compre- every domain, there are various domains where explainability
hensible. That is, it provides explanations on the pre- is demanded. Explainability is significant in healthcare do-
dictions made for individual data samples. It changes main to help the medical professionals to understand the un-
the individual data samples by altering the values of derlying reasons behind the diagnostic predictions provided
features and by analyzing the impact on the specific by the system. Banking sector also require explanations to be
output. This framework aims at training local surrogate provided to the user to understand why a particular credit was
models. The authors in [36] have made an analysis denied or to understand the reasons behind customer churn
on the advantages and the limitations of the LIME rate. Also, automobile industry will gain immense benefits
framework. The LIME provides a comparatively faster by adding explainability along with its decisions. There are
and simpler methods for locally explaining the black other domains such as agriculture, marketing, defense and a
box models. lot more that can benefit with the explainability that the AI
2) SHAP: SHAP is another XAI framework that is model provides. However, there are some challenges as well while
agnostic, post-hoc and can provide both local as well deploying the XAI-based systems.
as global explanations. SHAP stands for SHapley Ad-
ditive ex Planations(SHAP). This framework can pro- D. Motivation behind the integration of XAI and I5.0
vide explanations for tabular, text and image datasets. The integration of XAI with I5.0 shifts the emphasis from
Optimal credit allocations are connected with local mere automation to enhanced collaboration between humans
explanations here with the help of Shapely values. and the AI systems. The key benefits that arise from the
It is derived from the shapley values of the game integration of XAI in the context of I5.0 are highlighted

8 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

III. Applications of XAI in Industry 5.0

ge
w led s LIM
E
I5.0 aligns with the 9th and 11th sustainable develop-
o
Kn raph ment goals(SDG) proposed by the UN, namely Industry,
G Innovation and Infrastructure and Sustainable cities and

c
nosti
Communities to promote a sustainable transformation [43].

c
c

Mo L t-Ho
sti

Hoc
Post- al
el Ag
Ag al
no

Loc
This will promote a major reshaping of industrial processes

de oc
s
Po

Mod
l
with efficient information fusion. Therefore, this section

SHA
PIRL

bal presents the different applications of I5.0 with XAI. The inter
Glo pecific Model Agnostic
dependencies between various I5.0 applications with XAI

P
lS
Local, Global
de
Mo
Post-Hoc are drawn from a detailed analysis of (i) how the various
cutting-edge technologies enhance the specific application,
Mo
fic

d (ii) learning about the intricacies of data sharing, (iii) under-

Lo el A
st ca eci

ca gn
Po Lo Sp

Pr l, Glo ostic
-H l

standing workflows and (iv) aligning with the goals such as

oc
el

e-H b
od

oc al
Model Specific
M

Local, Global

SDGs, and human-centric collaboration.

FA
& Agnostic

Post-Hoc
P

C
LR

ET
S A. Smart Factory 5.0
Factual Smart factories employ cutting-edge technologies like AI,
Counter Robotics and digital twins to lower costs, boost productiv-
ity, and enhance product quality [44]. Smart factories can
FIGURE 2. XAI Frameworks make intelligent and self-organizing decisions in response to
changes in their surroundings. Also, they can communicate
with other manufacturers, suppliers, and clients to enhance
collaboration and gain business intelligence. 3D printing,
here. One of the primary advantages of XAI is transparency, autonomous robotics, ML, predictive analytics, Blockchain,
with the help of which human operators can make informed and cloud computing are some technologies employed in
choices based on a better understanding of the AI-generated smart factory deployment. Using these technologies, the fac-
recommendations. I5.0 focuses on improving collaboration tory may increase productivity and efficiency by providing
between humans and automated systems. XAI facilitates visibility in production. Also, intelligent factories can track
the understanding of AI decisions by humans, enhancing and monitor their machine operations, enabling better main-
the interaction between humans and machines. This allows tenance and repairs [45]. Smart factories intended to make
humans to have control and make accurate choices based on decisions about production scheduling, predictive mainte-
the recommendations provided by AI. Furthermore, offering nance, resource allocation, inventory control, and quality
understandable explanations to the user helps in building assurance. Decisions about data collection, analysis, and
trust between the humans and I5.0 systems. Trust plays utilization must also be considered to improve processes and
a pivotal role in the human-AI collaboration, particularly operations. Decisions should be made regarding integrating
in industries such as healthcare, finance, and autonomous various technologies, including cloud computing, augmented
systems. Humans can leverage the strengths of AI, thereby reality (AR), virtual reality (VR), and the Internet of Things.
fostering problem solving and increased overall productivity. Decision models can forecast future results and trends,
Accountability can be ensured for the systems, as XAI analyze data, and make choices on the most effective use
enables thorough audits. Explainability plays a crucial role of resources and employees. Predictive analytics, linear
in identifying and rectifying biases in the training data or programming, decision trees, and neural networks are some
algorithms. This, in turn, guarantees that AI applications of the decision models utilized in smart factories. These
strictly adhere to ethical standards. Furthermore, the inte- models can detect possible issues and provide solutions to
gration of XAI aligns with regulatory requirements and stan- increase productivity and lower expenses. Developing and
dards. Safety and reliability are of utmost important in vital supervising the factory’s optimal control systems can also
sectors such as manufacturing, healthcare, and transportation, be done using decision models, guaranteeing that production
while are important applications of I5.0. XAI guarantees the is done efficiently and effectively [13]. XAI can be used
transparency and trustworthiness of AI system operations, in smart factories to improve decision-making and ensure
which is essential in high-stakes environments where errors that robot assessments are well-informed and consistent with
can have substantial implications. In conclusion, the benefits predefined objectives. Multi-source information fusion in
of incorporating XAI in I5.0 are manifold. From improving I5.0 help in integrating data from diverse sources such as
decision-making, building trust, promoting collaboration, machines, AI systems, sensors and humans, thereby provid-
and tackling ethical issues, XAI plays a pivotal role in ing informed, precise and timely decisions. Based on the
shaping a future where cutting-edge technologies coexist context, data-level fusion, feature-level fusion, and decision-
harmoniously with human expertise. level methods could be utilized. Some of the commonly used

VOLUME , 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

models include JDL model, Bowman Df&Rm model, Luo- and explanations generation for appropriate stake holders of
Kay model and Pau model [46]. Providing context-aware I5.0 should not be overlooked.
decisions help in maintenance scheduling, quality control,
cost savings and improved sustainability. Robotic decisions
B. Healthcare 5.0
can be explained using XAI so that people can compre-
hend why a particular action was taken. Moreover, XAI Various cutting-edge digital technologies, such as
can locate and resolve production anomalies and enhance blockchain, IoT, cloud computing, and AI, enhance
process monitoring and management. Finally, explanations I5.0 in the Healthcare sector. Increased data accessibility,
that XAI can produce insights and suggestions for improving improved patient experience, and more individualized
manufacturing processes and operations [47]. care are the benefits of these technologies. Digital health
records, remote medical diagnosis and monitoring, virtual
medical consultations, telemedicine, medical compressors,
predictive healthcare analytics, and automated healthcare
1) Applications of XAI in smart factories systems are some I5.0 applications in the healthcare sector.
Smart factories can employ XAI to increase operational Healthcare providers can enhance the quality of care and
effectiveness and decrease downtime. In the smart factory, minimize expenses by utilizing the possibilities of these
XAI can assist in identifying and anticipating issues before technologies [15]. For the healthcare sector to successfully
they happen, reducing unplanned downtime and increasing adopt new technologies and procedures and maintain its
output. Moreover, XAI can help to identify areas in the competitiveness and efficacy, decision-making is essential in
factory that require improvement to increase the effectiveness I5.0. Decisions must be made regarding implementing the
of the manufacturing process, enabling the production of finest technology, policies, and infrastructure, the efficient
high-quality goods in a short period. use of data and the safety and security of patient data.
Predictive maintenance can reduce industrial expenses by One of the most crucial concerns with decision-making
forecasting faults and extending component lifespan. Fac- in healthcare is data sensitivity, and the decisions are
tories began monitoring their equipment recently, and most also life surveillance determining factors. Therefore XAI
data gathered are in proper working conditions. Increasing explanations must be understandable to patients of different
intrusion detection accuracy, spotting unexpected failures, literacy levels and other entities at stake in the healthcare
and adapting to changing operating and environmental con- sector. Instead of one-size-fits-all explanations, user or
ditions are significant obstacles to industry deployment. To machine-centric and dynamic explanations must be foreseen
combat this, AI-based predictive maintenance can be uti- in further XAI models for the healthcare sector. Healthcare
lized in industrial scenarios where semi-supervised predictive organizations must also decide the best approaches to
maintenance is required to locate and address unforeseen control costs, enhance patient support, and ensure access
issues. Fig.3 depicts the various benefits of Smart Factory to care. The healthcare sector must also decide how to
with the Adoption of XAI. [48]. The significance of anomaly effectively use the potential provided by digital healthcare,
detection for quality prediction has gradually evolved due to including how to best incorporate telemedicine and AI
increased data collection across various domains, including for medical support. To remain competitive and at the
smart factories. Due to data imbalance in the anomaly de- forefront of healthcare innovation, businesses must make
tection model, numerous efforts have been made to increase wise decisions in the healthcare sector of Industry, 5.0 [50].
discrimination efficiency in the current industrial process. Doctors provide personalized recommendations and real-
Anomaly detection is crucial because it plays a vital role time information using XAI, which might help patients
in predicting the quality of chromate processes, significantly better. Additionally, by offering insights into the courses of
impacting how complete they are. Additionally, collecting therapy that are most effective for every patient and assisting
image data for monitoring during production is challeng- in the identification of probable adverse events, XAI can be
ing, and prediction is problematic due to data imbalance. utilized to enhance patient safety. Patterns in patient records
To provide a more intelligible assessment of faults with can be identified by employing XAI, which can then be used
data from the chromate process, a model was developed to improve public health initiatives. Finally, XAI can assist
using an unsupervised learning-based Generative Adversarial healthcare decision-making by providing helpful insights to
Networks (GAN) model, which performs training with only medical professionals and healthcare systems [51]. Strategic
standard data images and adds a visual component to the Fast technology investment like identifying the technologies
Unsupervised Anomaly Detection with GAN (F-AnoGAN) that will have the most significant impact on patient
basis [49]. Thus, XAI has been applied in several applica- outcomes and operational efficiency can be adapted to
tions of smart factories to make proper decisions and increase manage the costs. Introducing the technologies gradually
production. Although various techniques were adopted to can help the healthcare organisations to manage costs and
accumulate information seamlessly, the challenges concerned risks. Telemedicine and AI can extend care to remote
with information fusion, like data quality for varied sources, and underserved areas, reducing geographic barriers.
should be foreseen. Consequently, liability upon integration AI-powered tools can improve diagnosis accuracy and

10 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

Industrial Augmented
Stakeholder specific decisions
Reality

Generate Decisions with

Unmanned
Cobot specific decisions
trucks

the help of XAI

Inputs for Smart

XAI Adaption
Explanations for decisions
Factory

IoT Sensors
made by cobots

Cobots Understandable decisions with

in operation explanations for entity in SCM

Predictive Machine Clear interpretable explanations

Analytics for productive maintenance

FIGURE 3. Benefits of XAI Adoption in Smart Factory

treatment recommendations, leading to better patient models and ensuring their accountability. The methods have
outcomes. By automating routine tasks and optimizing provided information about how the attributes lead to the
resource allocation, advanced technologies can reduce wait prognosis. They have not provided any information regarding
times for appointments and procedures. the features that help us to predict a particular patient’s class.
Integrated Gradients carry out this task. It is the only attribute
that specifies the features that induce the model to provide
1) Applications of XAI in Smart Healthcare sector a different forecast, called negative attributions, and the
XAI can be successfully utilized in medical diagnostics positive attributions assist in making a prediction. Thus, the
and therapy, patient monitoring, medical decision support, most important lesson gained from studying these methods
drug discovery, etc. XAI can enhance illness diagnosis and is that they all explain how different features contribute
therapy. Medical professionals can make more accurate and to the outputs of the model [53]. As the healthcare sector
fast diagnoses with AI algorithms, which can spot useful is sensitive, the data through information fusion should be
trends in healthcare data that are difficult for humans to dealt with carefully, focusing on data biases, time series
notice. AI can be utilized to provide individualized therapies information and interpretable explanations for the victims.
and treatments for patients. XAI can be used to keep track of Utilizing XAI with Healthcare 5.0 should not provide too
a patient’s health and spot changes, and it can notify medical much or too minimal information, as it may create unwanted
professionals when interventions are required. Doctors can chaos.
receive real-time suggestions and guidance regarding patient [54] presented a survey emphasizing how well XAI fits
care using XAI. AI algorithms can review patient data and into the healthcare 5.0 environment. The research developed
deliver immediate feedback on the most appropriate action. a classification and segmentation architecture for medical
XAI can be utilized to create more efficient clinical trial de- images using XAI. The architecture integrates Convolu-
signs. Clinical trial outcomes can be predicted, trial designs tional Neural Network (CNN)-based Deep Learning meth-
can be made more effective, and suitable trial volunteers can ods with Federated Transfer Learning (FTL) for detecting
be found using AI algorithms. Using XAI, drug discovery the Covid-19 disease. Raw data gathered from COVID-19-
and development can be accelerated. AI algorithms examine affected patients are efficiently normalized by the developed
vast amounts of data to pinpoint prospective therapeutic FTL-assisted CNN autoencoder classifier, which divides the
targets and forecast the efficacy of medication candidates dataset into five classes. The system evaluates the classifier’s
[52]. prediction and provides the decision using the proposed
Human life depends on clinical diagnosis models, so we explainability module and the explainable diagnostic module.
must have the confidence to handle a patient as a black-box Hence, XAI can be used in various healthcare system appli-
model directs. A study is done, and it provides examples cations to provide better patient support and assist medical
drawn from the heart disease dataset and explains why ex- practitioners. XAI can significantly contribute to public
plainability approaches must be chosen when utilizing deep health initiatives by providing insights into complex patterns
learning methods in the medical field. The most common within patient records. By understanding these patterns,
feature-based techniques, such as LIME and SHAP, have healthcare organizations can identify risk factors, develop
been discussed in the study. Although they serve relatively targeted interventions, and improve overall outcomes.
similar purposes, they are approached quite differently. These
are crucial for exposing the opaque nature of black box

VOLUME , 11

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

C. Governance 5.0 tegrity. Although various reference architectures for Industry

I5.0, the next Industrial Revolution, collaborates humans 4.0 were available to address the technology adoption and
with machines, enabling advanced manufacturing and pro- usage concerns, one baseline architecture is required for I5.0
duction capabilities by integrating digital, physical, and [57]. One such architecture for Industry 4.0 is evaluated by
biological technologies. As such, it is essential to ensure proposing a new reference architecture in detail to promote
that efforts to build and deploy new technologies are made its usage for I5.0 in bringing humans in the loop for secure
responsibly and with a consideration of the implications for AI in smart manufacturing [58].
global governance. Good governance in I5.0 is essential to Most of the issues concerning the usage of sensitive data
ensure that these new technologies are used for the benefit of in AI have been alleviated by XAI adoption as it provides
society, not for the help of a few [13] and assure improved insights into the black box of AI. It also helped satisfy
business agility. To overcome this situation, the government the data privacy policy enforced by the European Union‘s
and other private organizations should join hands to craft General Data Protection Act [59], [60]. XAI can help to
various policies and required regulations that enforce public improve the governance of AI systems by providing greater
protection and nurture liable innovation. This includes pro- transparency and understanding of how decisions are made
moting data privacy and security, ensuring the ethical use of and why. XAI can help ensure that decisions made by AI
AI, and developing guidelines for using smart manufacturing systems are logical, fair, unbiased, and based on accurate and
technologies. In addition, various governments and industrial up-to-date data [61]. XAI can also be used to help identify
organizations should focus on creating a universal framework potential areas of risk or areas of improvement that can be
that summarizes the values of noble governance in the addressed to improve the governance of AI systems. XAI
industry practices [44]. The integration of XAI with I5.0 can also evaluate the AI system‘s performance and identify
poses various challenges, especially in the context of smart areas for additional training or other interventions. XAI
manufacturing. However, these challenges could influence can provide comprehensive justifications for predictions and
several future research directions in this domain. One of guarantee that people comprehend the decisions made by AI
the significant areas is that of human-centric interactions algorithms.
[55]. It becomes mandatory to design intuitive and context-
aware frameworks and interfaces. Similarly, evaluation met- D. Transportation 5.0
rics for evaluating the trustworthiness of a system is of Smart transportation in I5.0 focuses on creating autonomous
utmost importance. Developing light weight and real-time technologies and systems to improve the effectiveness of
XAI systems which are scalable is another critical area of transportation networks, including advances in intelligent
research. Sustainability is one of the major focus areas of traffic systems, automated driving, and digital tools to
I5.0. Hence, green manufacturing could be studied for future monitor and manage traffic flow. Autonomous vehicles are
research. anticipated to become more prevalent, using cutting-edge
Furthermore, for efficacious governance, decision-making sensors and AI to enable them to drive safely and effectively.
with appropriate and adequate user-specific explanations for To improve traffic planning, real-time data gathering and
various decisions made in I5.0 applications is mandated. As analysis will be made possible by cutting-edge technology,
such, decision-making needs to be done in a more informed such as 5G networks. Blockchain-based solutions will also
and data-driven manner. This requires advanced analytics assist in protecting data and guarantee its accuracy. By
to make informed decisions and automated decision-making improving routes and easing congestion, smart transportation
systems to help with real-time decision-making. Addition- systems will also significantly reduce emissions [62].
ally, decision-makers need to be able to access and analyze Features of smart transportation systems are
data to make informed decisions promptly and quickly. Ulti-
1) Autonomous Vehicles: In I5.0, autonomous vehicles
mately, decision-making in I5.0 requires an agile approach in
are predicted to transform the transportation sector
a rapidly changing environment like Governance and policies
completely. Advanced technologies like ML, computer
must be curated through careful examination and evaluation.
vision, and AI will empower autonomous vehicles
[63]. Autonomous vehicles can avoid obstructions and
other traffic while spotting patterns and making deci-
1) Applications of XAI in Governance in Industry 5.0 sions based on the information they gather [64].
Various frameworks for realizing the ethical principles were 2) Smart Traffic Control: I5.0 is anticipated to introduce
suggested in the literature, such as responsible data design, smart traffic control technologies. These systems iden-
a process-based framework for governance, and actionable tify traffic patterns and modify signal timings based
principles [56]. Furthermore, these rich manufacturing use on those patterns using sensors, cameras, and AI algo-
cases and further adoption of this I5.0 in various sectors rithms. This will aid in reducing congestion in traffic
require standards for building open components compatible and improving transportation efficiency [65]. To secure
with all other parts for effective integration. Furthermore, it IoT data in smart traffic control systems, measures
can provide exploratory reports on monitoring business in- include encryption, authentication, regular updates,

12 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

network segmentation, and intrusion detection systems. 1) Applications of XAI in Smart Transportation System
Blockchain enhances data integrity through its de- There are numerous ways that XAI can be applied to enhance
centralized structure, immutability, transparency, con- the smart transportation system. Better traffic flow and
sensus mechanisms, and smart contracts. Blockchain vehicle routing forecasts can be made with XAI, resulting in
architecture prevents data tampering, ensures transpar- more effective transportation systems. XAI can also generate
ent transaction verification, and automates processes more precise predictions of travel duration and delays to
through self-executing contracts. improve traffic management and safety. Additionally, XAI
3) Connected Infrastructure: I5.0 will heavily rely on can more accurately detect and forecast traffic accidents and
connected infrastructure. It will enable the Internet of other occurrences, enhancing safety and easing congestion.
Things to be connected to transportation networks, pro- Better visualization tools like 3D maps and AR, which
viding real-time information about traffic congestion can aid drivers and passengers in better understanding their
and other circumstances. As a result, transportation surroundings and making better judgments, can also be
authorities can take the initiative to boost the effec- provided by XAI [69].
tiveness of transportation networks [66]. XAI improves the transparency of decision-making in
4) Smart Ticketing: In I5.0, smart ticketing systems will autonomous vehicles by providing detailed explanations of
increase transportation productivity. Passengers can how and why decisions are made, particularly in crucial
buy tickets online and pay for transportation using their circumstances. The transparency fosters public confidence,
mobile devices. As a result, fewer physical tickets will guarantees adherence to regulatory norms, and enables ongo-
be required, and buying tickets will be significantly ing enhancement of driving algorithms. During an unavoid-
simpler overall [67]. able crash scenario, XAI explains the justification behind
5) Sustainable mobility: I5.0 is anticipated to focus sig- the vehicle’s selection of a particular path. XAI plays a
nificantly on sustainable mobility. This will include crucial role in providing auditable decision trails, which are
employing electric vehicles to reduce the carbon im- necessary for legal and insurance reasons. This facilitates
pact of transportation and employing renewable energy effective communication with all parties involved. [70].
sources like solar and wind to power automobiles [68]. Utilizing a multimodal deep learning architecture, traffic
Various challenges of Transportation 5.0 and benefits flow prediction employing XAI can assist in predicting traffic
of XAI in uplifting or mitigating those challenges in patterns, allowing transportation planners to create plans to
transportation 5.0 are depicted in Fig. 4 ease congestion and boost safety [71]. Route Optimization
using XAI can optimize routes for different transportation
modes, such as public transit, cars, and buses, to reduce
travel time and improve efficiency [72]. XAI for Ridesharing
can match passengers with drivers and optimize routes to
Smart transportation in I5.0 is mainly about integrating
reduce wait times and improve the overall passenger expe-
advanced technologies such as AR, Big data, IoT, and AI
rience [73]. Public Transport Management using XAI can
into the transportation system to improve efficiency and
manage public transportation fleets, helping reduce delays
decrease costs. The decision-making in this domain will
and improve efficiency [74].
involve integrating these technologies with the existing trans-
portation systems to optimize operations. Decisions like load
optimization, path selection, predictive maintenance, traffic
flow optimization, and others fall under this category. E. Education 5.0
The decisions will be generated from data analysis, AI Data-driven, personalized, and hands-on training are the
algorithms, and data gathered from multiple sensors. The primary objectives of smart education in I5.0. It will in-
decisions adopted should minimize transportation costs, in- tegrate established instructional strategies with cutting-edge
crease system effectiveness, diminish the adverse impacts of technologies like VR, AI, AR, and ML. It will be used to
traffic on the environment, and guarantee the well-being of acquire the abilities and information required to perform in
the passengers [11]. the digital sector. Smart Education would employ AI to offer
Smart transportation requires decision-making to optimize customized learning paths, ML to provide individualized
the flow of people and goods. This involves making choices feedback, AR/VR to deliver immersive learning experiences,
on deploying resources like trains, buses, and ride-sharing and smart analytics to monitor the progress.
services, as well as positioning traffic lights and road signs. Education 5.0 is an idea of digital transformation in
AI systems that use information from cameras, sensors, and teaching and learning that uses the most recent innova-
other sources to discover patterns and foresee probable traffic tions and market trends to produce a more individualized,
disturbances are frequently used to make these judgments. engaging, and successful educational experience. Big data
The choices taken by these algorithms can assist in increas- analytics, AI-powered learning systems, VR, and AR are all
ing safety, decreasing congestion, and improving traffic flow examples of technology that could cater for the learner’s
[62]. needs. Education 5.0 in I5.0 improves learning and teaching

VOLUME , 13

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

Transportation 5.0
XAI in advancing
Encouraging
Sustainable Practices
Improving Customer
Experiences
Personalized Travel
Reducing Emissions
Real-Time Information
Resource Optimization
Enhanced Interaction
Promoting Eco-Friendly Modes

Optimizing Logistics
and Efficiency

Fostering Innovation
Route Optimization and Development
Load Balancing
Energy Efficiency
Enhancing Safety and
Reliability Research and Development
Adapting to New Technologies
Collaborative Solutions

Predictive Maintenance Facilitating Regulatory

Accident Analysis Compliance and Ethics
Trust in Autonomous Vehicles

Compliance with Regulations

Ethical Decision Making
Auditing and Reporting

Integration of Sustainability and

Data Management Regulatory and Safety and
Advanced Environmental
and Privacy Ethical Issues Reliability
Technologies Impact

Challenges of Transportation 5.0

FIGURE 4. Challenges in Transportation 5.0 and XAI in advancing Transportation 5.0

environments by reducing the space between industry and more interesting and productive learning experience with the
academics [75]. aid of AI [76]. Various issues in Education 5.0 incorporation,
AI in education can significantly change how students advantages of XAI and its benefits in Education 5.0 is
learn in the classroom. A customized learning plan is de- depicted in Fig. 5.
veloped for each student based on their interests and needs
when deploying AI to personalize education. AI can also
automate the grading process, freeing teachers’ time to focus 1) Applications of XAI in Education 5.0
on instruction and student interaction. AI can also be utilized The application of XAI in education can completely trans-
to design individualized learning environments and virtual form the way students learn. XAI improves personalized
classrooms. In addition, AI can assist with data analysis and education by analyzing the unique learning styles, strengths
research, enabling teachers to make better-informed choices of individuals. It customizes content by comprehending
about how to teach their kids. Teachers can give pupils a students’ requirements through their interactions, adapting
learning paths, and modifying content hardness in real-

14 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

Education 5.0 Issues in Education 5.0

Detailed explanations for student
1
skill-specific understanding

Advantages of XAI for

ion nt
rat nte
Interpretable Marking & results

Education 5.0
2

Lo
ne g co
analysis

ss
in

o
ad

fJ
Recommended areas for
3

ob
sle
ge
improvement

Mi
Privacy on new 4 Captivating learning experience
methodologies created

Skill-based Feedback based evaluations Notified tutoring

personalized training and subsequent learning system

1 2 3 4 5
Immersive learning Exploratory programs
experiences monitoring

Benefits of Education 5.0

FIGURE 5. Issues in Education 5.0 and leveraging the XAI for better learning

time. XAI presents explicit feedback and suggestions while made possible by XAI. Based on each student’s unique
identifying motivational factors to enhance user participa- requirements and preferences, learning experiences can be
tion. This facilitates the establishment of a flexible and tailored using AI-based algorithms [79]. Each student’s re-
adaptable learning environment and enhances transparency quirements can be satisfied by creating courses with XAI.
in the educational process, enabling students and educators Course content customized to each student’s requirements
to better understand the explanations behind customized and interests can be created using AI-based algorithms.
modifications and choices. Additionally, they are employed to offer specialized student
XAI can assist in locating student learning gaps and offer- support. Students can receive individualized support from
ing solutions to fill them. Additionally, instructors can utilize XAI through timely reminders and notifications, customized
XAI to determine the most effective ways to teach specific resources, and feedback. Students are given interesting con-
ideas and better understand their pupils’ strengths and limi- tent, such as interactive learning modules, games, and sim-
tations. Additionally, by measuring learning outcomes with ulations, utilizing AI-based algorithms [80]. Inappropriate
XAI, educators may better understand what works and what information generation, more curated contents, provoking
doesn’t in the classroom [77]. XAI helps analyse student data should be monitored while tailoring XAI applications
performance data to precisely identify specific points where for Education 5.0
the student is struggling. Knowledge derived from XAI
aids tutors in adjusting their approach to individual needs F. Agriculture 5.0
based on the advancement of each learner. XAI recommend ”Agriculture 5.0” refers to applying cutting-edge technolo-
applicable educational resources or methods applied to other gies to enhance agricultural output and boost farming pro-
similar students that have shown success. XAI continuously ductivity. It uses sensors, robots, drones, big data, and IoT to
analyses how students respond to different techniques and increase agriculture yields and animal husbandry. The goal
learning materials. XAI improves communication between behind the agricultural revolution known as ”Agriculture 5.0”
teachers and students by providing appropriate, correct, and is to use cutting-edge technologies, including automation,
precise evidence-based justifications for academic decisions robotics, IoT, AI, and big data, to increase the productivity
and educational improvements. and sustainability of agricultural processes. This revolution
Learning analytic methodologies for student assessment aims to make global food production more effective and
can be automated using XAI. To accurately analyze student sustainable. Improvements in agricultural inputs, precision
performance on tests and other assessments, AI-based algo- agriculture technologies and procedures, and the incorpora-
rithms are deployed [78]. Research on intelligent tutoring tion of agricultural data into decision-making processes are
systems (ITS) focuses on creating instructional strategies to all part of it. Agriculture 5.0 aims to bring about a more
replicate students’ fundamental needs, circumstances, and efficient and sustainable agricultural revolution that will also
talents, such as subject expertise, meta-cognitive abilities, increase employment and improve people’s quality of life
and mental states, and provide individualized instruction. [81]. AI in agriculture 5.0 applies AI tools like computer
Students can benefit from personalized learning opportunities vision, big data, and ML to enhance crop production and

VOLUME , 15

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

farm operations. AI can help farmers manage resources more 1) Applications of XAI in Agriculture 5.0
effectively, automate labor-intensive operations, and enhance Several research works have been done by applying ML
agricultural yield [82]. algorithms and AI techniques for decision-making, such as
Issues in agriculture prediction and classification in the agriculture domain. In
those works, it is unclear how the learning process is done
1) Crop selection: The identification of suitable crops and the rationale behind the decision-making. To solve this
to be cultivated depends on regional circumstances, issue, an XAI has been applied to the agriculture dataset,
market trends, and consumer preferences. including soil, weather and other parameters pertaining to
2) Irrigation Management: Choosing when and how much crop yield for data analysis. The data analysis provided
water to apply to crops and determining how much the decision that no-tillage management can enhance the
crop productivity would be affected by water scarcity. yield of the maize crop. The XAI and interpretable ML
3) Fertiliser application: Choosing the right fertilizers, algorithms have provided the answers, such as the useful
how much to apply, and how to minimize the envi- parameters important for prediction, the association between
ronmental impact. input and response parameters and the reason behind the
4) Weed control: Identifying the appropriate weeds to predicted value for a particular observation. Further, it also
remove and figuring out how to deal with them, provided information on whether other ML algorithms have
including applying herbicides, crop rotation, and other provided the same answer. The XAI has proved that the
cultural practices. questions were unanswered by other ML algorithms but
5) Pest Management: Choosing the best pest control XAI and interpretable ML can enhance the explainabiity
procedures, including the use of pesticides and other and trust in AI [86]. The explainability nature has been
pest control techniques, that cause the least amount of explored by applying XAI to the dataset of wheat head
disruption [83]–[85]. detection. The dataset has information collected from various
repositories related to different growth stages and lighting
XAI can be utilized to develop decision models to manage conditions. It is challenging to analyze such data, leading
the issues and improve crop production. Leveraging the to poor model stereotypes. To solve this issue, a pre-trained
power of XAI in agriculture 5.0 and its benefits are depicted YOLO model was applied to detect wheat heads using the
in Fig. 6. SHAP library. In the developed model, SHAP helps to detect
a partially overlapped wheat head. The model proved to have
the ability to minimize the cost and time in the supervision
of wheat crops [87]. An XAI framework with ontology
Enhanced Decision Making design and knowledge map model was proposed for digital
Clarity in Complex Data
Trust and Reliability agriculture to provide human-understandable explanations
Informed Choices
Predictive Analysis User Confidence for the decision and results. The framework was developed
Transparency
Accountability to provide algorithms, definitions and information for devel-
oping data mining-based models. Further, Agriculture also
Customization and Precision
Agriculture
proposed an agriculture computing ontology for explaining
Learning and Improvement
the knowledge mined on various datasets [88]. A system
Tailored Solutions
Model Refinement
Efficiency in Resource Use
Continuous Feedback Loop
was developed to detect plant disease using XAI techniques.
Crop Health Monitoring
Adaptive Learning
In the model, the XAI method, namely Gradient weighted
class activation mapping++, was applied to identify plant
Collaboration and Interdisciplinary
disease and emphasise the regions covering the leaves that
Integration lead to classification [89]. Storage, security, reliable and
Cross Disciplinary Understanding
Risk Management accurate results are some of the factors to be considered
Joint Problem-Solving
Innovation through Diversity Identifying Risks in the evaluation of XAI for Agriculture 5.0.
Strategic Planning
Proactive Measures

G. Energy 5.0
In I5.0, ”Smart Grid” refers to integrating cutting-edge tech-
Agriculture 5.0 With XAI
nology into the electricity grid to improve its effectiveness
and efficiency. This includes adopting smart meters, sensors,
and distributed energy resources (DERs) to facilitate a two-
way flow of electricity from utilities to end consumers. In ad-
dition to using analytics, automation, AI, and IoT, the smart
FIGURE 6. XAI Adoption in Agriculture 5.0
grid in I5.0 also uses these technologies to improve resource
efficiency and control the distribution and consumption of
electricity. The ultimate objective of the Smart Grid in I5.0 is

16 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

to lower prices, boost dependability, and aid in the reduction comprehend the reasoning behind the system’s judgments
of greenhouse gas emissions. The larger power consumers if we hope to increase user confidence in these systems.
of today are industrial sectors who started to use the SG Additionally, XAI can assist operators in spotting abnormal
even decades ago before knowing the term as a whole. The behavior and modify the system as needed. Additionally,
major concerns with today‘s energy systems are greenhouse XAI can offer perceptions into how the system operates and
gas emissions, the economics involved in rapid hikes in assist operators in anticipating potential issues before they
electricity prices, worsened grid reliability, and weakened become serious. Finally, XAI can enable automated systems
energy security (When electricity is transmitted via electric that can react swiftly to modifications in system behaviour to
vehicles or other plug-in vehicles) apart from the traditional maintain systems’ smooth operation and minimize hazards.
electric grid‘s issues such as lack of storage capabilities, volt- Challenges in Energy 5.0 and services of XAI in Energy 5.0
age level distribution, interrupted power service, and internal are depicted in Fig. 7.
equipment failure with complex wiring. The consumer‘s role
is consuming and producing electricity by sharing the energy
when required, thereby termed a prosumer. The concept 1) Applications of XAI in Energy 5.0
of Energy 5.0, which encompasses the various industrial XAI contributes to numerous Energy 5.0 solutions. Anomaly
interactions with energy development [90], enforces human detection is one of them, where XAI may be used to
intelligence in the modeling and analysing smart energy find abnormalities in the smart grid and notify operators
systems. if anything is not working properly. This can help reduce
Though the smart grid offers effective solutions for power downtime and improve system reliability. XAI for predic-
management by moving operations and decision-making to tive maintenance can also predict when certain smart grid
the source of power generation, some of the major issues components need maintenance or replacement. This can
are: help reduce costs and improve the efficiency of the system
[92]. XAI in Load Forecasting can be used to forecast the
1) Cybersecurity: As grid systems become increasingly demand for electricity and help optimize the use of resources
interconnected and automated, the potential for mali- [93]. Demand Response using XAI can accurately predict
cious cyber activity increases. the electricity demand and help optimize resource use of
2) Data privacy: As more and more data is collected resources [94]. This can help reduce costs and improve
from smart meters and other sources, it‘s important to the system’s efficiency. Furthermore, XAI for Security and
ensure that data is collected, stored, and used securely. Privacy can detect potential threats and vulnerabilities in the
Data privacy of sensitive information about data usage, smart grid. It has been used for intrusion detection in the
and sharing is also a major concern. smart grids [95] with various algorithms and their usage
3) Network reliability: With so many connected devices, for intrusion detection. This can help reduce the risk of
it‘s important to ensure that the grid remains opera- malicious attacks and improve the security and privacy of
tional and resilient in the face of physical or cyber the system.
threats.
4) Cost: While many technologies that make up the smart
H. Summary of applications
grid are cost-effective, the upfront costs of deploying
the infrastructure can be high and require long-term The different verticals of I5.0 were discussed with the
investments. application of XAI. These include smart Factors, healthcare,
5) Regulation: SG technologies are complex and, in governance, transportation, education, agriculture 5.0, and
some cases, may require new regulatory frameworks energy 5.0. I5.0 in all these verticals was discussed with the
in order to ensure their safe and effective use. employability of XAI in generating user-specific explana-
tions by considering different entities at stake. Utilizing XAI
XAI is required in smart grids because it allows decision- models in these verticals will enhance the performance and
makers to better understand the system‘s decisions. expose some open issues, namely, security and privacy, in-
Decision-makers can use XAI’s practical insights to assist teroperability, transparency, and lack of regulatory concerns
them in making decisions that are consistent with their in adoption. XAI’s models can explain the decisions based
goals and objectives. With the assistance of XAI, decision- on different users’ needs and perspectives. This, in turn, may
makers will be able to address potential issues and obstacles help ensure trust in adopting AI-based I5.0 applications to
in the smart grid system more quickly. By offering more the fullest.
insight into the underlying data and operational processes,
XAI in smart grids can enhance decision-making processes. IV. Open Issues, Challenges and Future Directions
Engineers and operational staff can benefit from XAI by XAI is projected to play a significant part in I5.0, such
making faster, more accurate decisions that improve effi- as manufacturing, logistics, and healthcare, combining AI,
ciency and reduce costs [91]. Smart grids employ XAI to de- IoT, and automation. Although many research works have
fend decisions made by AI-enabled technologies. Users must been proposed for XAI, the intricate and black box features

VOLUME , 17

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

Energy 5.0

Explanatory energy sharing among

Micro grid prosumers
1

Recommender
Energy production
Exploratory Energy control &

system
center
tracking
2

Services of XAI
Explanatory electricity bills on energy
consumption
3

Interpretable predictions on future

energy production
4

Smart meters Stakeholder specific explanations on

Electric energy supply chain
vehicles 5

Cyber security Network reliability Maintenance cost Lack of regulatory laws Lack of explanations

1 2 3 4 5

Challenges of Energy 5.0

FIGURE 7. Issues in Energy 5.0 and XAI Services in Energy 5.0

of XAI raise some open issues in I5.0-related applications. customized explanations pertaining to individuals as
Table. 3, explains the open issues and challenges face the they can highlight the explainability factors related to
integration of XAI within I5.0. Addressing these challenges specific user preferences and their responsibilities.
through the proposed future directions significantly enhance 3) Trust-worthy explanations for complex models:
the impact and reliability of AI systems in industrial appli- XAI also helps improve the models’ trustworthiness.
cations. Currently, deep learning models are widely used for
I5.0 applications. Although their responses are quite
1) Explainability-by-Design: The applicability of XAI opaque, still integrating XAI could help in improving
is presently available for existing models. The role of the trustworthiness, simplicity, and accuracy by ana-
incorporating explainability principles into a model is lyzing the explanations.
to ensure enhanced transparency from the beginning 4) Real-time explainability: Industrial applications are
while retaining the model’s performance. Integrat- not only memory-constrained due to the extensive
ing the explainability elements into the intrinsic AI use of edge devices but also time-constrained as the
training and development process can provide more operations need to be performed in real-time. There-
straightforward explanations regarding the model per- fore, XAI methods need to be designed so that they
formance. are computed in real-time to comply with the timely
2) Context-aware explanations: Explanations about a decision process.
specific industry sector is essential to highlight the 5) Integration with diverse data formats: I5.0 considers
unique intricacies associated with it. It helps in dealing many diverse data types that can be obtained from
with essential customization so that the dynamics of sensors, cameras, or any other device. This results in
human-machine collaboration, safety standards, and multimodal data types such as images, text, audio, and
operational objectives are improved. The effectiveness other information formats. Specialized XAI methods
of Human-machine collaboration also benefits from the

18 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

must be created so that the data types can be integrated nological, decision-making, and organizational conse-
and explained within the same framework accordingly. quences and implications. This survey expands the de-
6) Human-centered interaction and trust: The in- sign area for Human-Centered XAI techniques. Some
tegration of human-machine collaboration within the XAI systems may deliver simplistic explanations that
industrial environment is based on interfaces through cannot represent decision-making complexity and un-
which humans communicate to the machines and the dermine system trust. Based on this new problem,
trust that the machines would do the job appropriately. Tubella et al. [99] proposed a mechanism for evalu-
Both tasks, i.e. interfaces and trust, can be improved ating AI moral limits based on inputs and outcomes.
by analyzing the XAI explanations. However, XAI Provable rules that control system inputs and outputs
techniques should be designed to cater for both the are presented with clear moral principles. This tech-
interfaces and trust while explaining the model. nique increases explainability by allowing stakeholders
7) Standardization and Evaluation: XAI is an emerging to review the system’s interpretation and use of moral
field and so does I5.0. There is a lack of standard- principles. Focusing on inputs and outcomes helps
ization and evaluation metrics in the domain of XAI evaluate AI systems and examine their morality. This
and I5.0. Therefore, there is a significant demand strategy improves AI system users’ confidence and
for the development of standards and metrics that well-being by clarifying its moral standards. In some
would provide a guideline for using such emerging XAI systems, the underlying data required to train the
technologies and a standard way to validate them. model may be private or inaccessible. This situation
The explanations provided by the XAI models are makes it hard to evaluate the model’s correctness or
evaluated using two different ways: subjective and comprehend data biases. Winter et al. [100] examined
objective. In the case of the subjective metric, domain the limited acceptability of AI and ML technologies
experts’ input and the random audience input are used in hospitals due to a lack of epistemic openness.
for evaluation. In terms of objective metrics, metrics The authors discussed how an AI application for the
are devised specific to the task, such as contrafactual early identification of a rare respiratory ailment was
and example-based methods and, on the other hand, developed and how tight collaboration between AI
based on models like model trustworthiness and at- engineers, physicians, and biological experts might
tribution. Some commonly used objective metrics are improve transparency. Overall, transparency is a sig-
Simplicity, sensitivity, completeness, stability, compu- nificant issue in the XAI-related I5.0 applications. In
tational cost, monotonicity, correctness of explanation, the future, more appropriate countermeasures should
confidence of explanation, fidelity, generalizability and strengthen openness to enable comprehension and con-
sparsity. Some subjective metrics are comprehensibil- fidence while still being practical and efficient for XAI.
ity, satisfaction, efficiency, pervasiveness, trust, effec- 10) Interpretability: Even with substantial openness, XAI
tiveness, justifiability, interestingness, goodness and may not be sufficient to comprehend an AI model fully.
transparency. However, there is no proper standard This can make it difficult for consumers to trust the
evidence for the implementation of these metrics [96]. model’s conclusions and for regulators to determine
8) Ethical considerations and regulations: Security if it is compliant. XAI depends on humans’ capacity
and privacy of individual and industrial information to comprehend and explain the behavior of an AI
is one of the utmost priorities for human-machine system. A standardized definition of interpretability
collaboration to work. XAI explains, however, that makes comparing and evaluating XAI solutions im-
these explanations can be compromised and manip- possible. As AI models become increasingly integrated
ulated through various attacks. Some of the potential into decision-making processes, consumers and stake-
risks or ethical concerns that might arise with wider holders must be able to rely on the model’s output.
adoption of XAI in Industry 5.0 applications include Interpretability enables users to comprehend how the
bias, misinterpretation of explanations, accountability, model arrived at its conclusions, fostering trust and
and excessive reliance on AI [97]. In this regard, XAI ensuring that the AI system’s decisions can be held
methods should be designed to comply with ethical accountable. In [101], the authors stated transparency
concerns and data privacy rules. and accountability in AI systems would help peo-
9) Transparency: In I5.0, some XAI systems may give ple understand AI algorithm decisions. In particular,
transparency, but not all the knowledge is included to black-box AI and algorithmic decision-making are
comprehend a decision or prediction that might make increasingly challenged. Their study analyzes how AI
an unconvinced system for users. For the transparency explainability affects user trust and attitudes toward
issue in the XAI, Ehsan et al. [98] introduced Social AI. It conceptualizes causability as an antecedent of
Transparency (ST), a viewpoint on XAI that considers explainability and a vital cue of an algorithm. It
socio-organizational context. The authors interviewed analyzes how they affect the user-perceived perfor-
AI users and practitioners to investigate ST’s tech- mance of AI-driven services concerning trust. More-

VOLUME , 19

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

over, interpretable models allow developers to easily Users’ comprehension and trust in ML models is
find and fix AI system biases and faults, improving crucial to their successful use in cybersecurity. Trans-
prediction accuracy. Deep learning models are used parency and explainability increase as black-box ML
in application like temporal multilabel classification algorithms make critical predictions. Cybersecurity
(TMLC) models, which assign multiple labels to input professionals require explanations for ML model out-
sequences, like video activity recognition and anno- comes.
tation. Debugging TMLC models is challenging due Recent research has focused on explainability ap-
to the exponential growth of label combinations and proaches, white-box interpreter assaults, and explana-
the need to understand how errors propagate within tion characteristics and metrics. Security characteris-
sequences. Researchers have developed DETOXER tics, cybersecurity threat models, and black-box as-
[102], an explainable, interactive visual analytics tool saults on explainable models are missing. Kuppa et al.
for debugging video activity recognition models to [104] presented an XAI taxonomy for cybersecurity-
address these challenges. DETOXER provides multiple related security features and threat models. A new
scopes of explanations and outputs, including frame- black-box attack analyses gradient-based XAI algo-
level outputs, video-level explanations, and global- rithms’ consistency, accuracy, and confidence secu-
level explanations. This helps users detect errors of rity. The developed system misleads the classifier and
various scopes and types through exploration and explanation report on three security-relevant datasets
interactive visualization. In certain fields, including and models without impacting the classifier output.
healthcare and finance, regulations require explana- These insights can improve XAI security and ro-
tions for AI-driven decisions. The ability to interpret bustness. Moreover, explainable approaches boost the
aids in meeting these requirements and ensuring ethical deployment of ML methods, which combat growing
compliance. The authors in [103] announced that AI cybersecurity risks. Counterfactual explanations are
systems must comply with human rights, the law, and popular because they assist users in comprehending
the public welfare. To maintain public safety, social black-box model decisions and identifying mutually
stability, and innovation, governments and politicians exclusive data examples that would influence out-
must develop real policies within moral, legal, and comes. Recent XAI research has focused on explain-
cultural frameworks. Well-established tools and design ability methodologies, white-box interpreter assaults,
techniques can assist governance in reducing nega- and explanation characteristics and metrics. Model
tive episodes, developing trust, and stabilizing society. explanations can add attack surfaces to underlying sys-
Regulations that encourage socio-legal and technical tems, although research is few. Membership inference,
compliance can boost innovation. The policy must as- model extraction, poisoning, and backdoor attacks
sign human accountability for AI system development can compromise system privacy using explanations.
and deployment to fill decision-making gaps caused by This research examines counterfactual cybersecurity
automation. Regulating AI helps promote well-being features and threat models to fill the gap. Based on the
for everybody in a sustainable world. XAI can help above analysis, Kuppa et al. [105] suggested a black-
create ethical and law-abiding AI systems that benefit box XAI attack to violate classifier confidentiality and
healthcare and finance. XAI in policymaking ensures privacy. The approach is evaluated with cybersecurity
AI systems align with human values, encourages trust, domain datasets and models, showing its effectiveness
and promotes openness, creating a more responsible under real-world threat models and emphasizing XAI
and beneficial AI ecosystem. Future directions for security. Nguyen et al. [106] developed a conceptual
the interpretability of XAI will focus on achieving a framework called ”Human-in-the-Loop XAI-Enabled
balance between interpretability and performance in AI Vulnerability Detection, Investigation, and Mitigation”
models and developing techniques that adapt to user (HXAI-VDIM) is proposed. HXAI-VDIM integrates
requirements, scale efficiently, and address domain- security analysts or forensic investigators into the man-
specific challenges. This will enhance the comprehen- machine loop, leveraging XAI to combine AI and In-
sion, credibility, and adoption of AI across various telligence Assistant (IA) to amplify human Intelligence
applications. for proactive and reactive processes. HXAI-VDIM
11) Security: Other than transparency and interoperability, aims to create an interactive, iterative loop between
security is another issue that should be considered to humans and machines using security visualization,
protect sensitive data and prevent AI system. Secure enabling human Intelligence to guide the XAI-enabled
XAI solutions protect privacy, user trust, and AI-driven system and generate refined solutions, emphasizing the
judgments while providing explanations. Security also importance of security in XAI. Future directions for
reduces the possibility of adversarial assaults on AI XAI security will emphasize augmenting robustness,
models that could mislead and hurt. privacy, and dependability while preserving explain-
ability. This includes developing secure explanation

20 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

methods, addressing potential attack surfaces, imple- REFERENCES

menting privacy-preserving techniques, and developing [1] Y. Lu, H. Zheng, S. Chand, W. Xia, Z. Liu, X. Xu, L. Wang,
human-machine integrated frameworks to foster user Z. Qin, and J. Bao, “Outlook on human-centric manufacturing
towards industry 5.0,” Journal of Manufacturing Systems, vol. 62,
trust and safeguard sensitive data. Quantum computing pp. 612–627, 2022.
could further enhance XAI’s model interpretability, [2] S. Ramakrishna, A. Ngowi, H. D. Jager, and B. O. Awuzie, “Emerg-
which would excel in extracting relevant features from ing industrial revolution: Symbiosis of industry 4.0 and circular
economy: The role of universities,” Science, Technology and Society,
complex, fused data and can easily interpret data of vol. 25, no. 3, pp. 505–525, 2020.
different dimensionality more effectively. Further, it [3] M. L. Smith, L. N. Smith, and M. F. Hansen, “The quiet revolution
could assist in generating counterfactual explanations in machine vision-a state-of-the-art survey paper, including historical
review, perspectives, and future directions,” Computers in Industry,
through quantum mechanics [107]. Similarly, emerging vol. 130, p. 103472, 2021.
technologies like brain-inspired computing or Neuro- [4] A. Jimeno-Morenilla, P. Azariadis, R. Molina-Carmona, S. Kyratzi,
morphic systems could play a vital role in I5.0 with and V. Moulianitis, “Technology enablers for the implementation
of industry 4.0 to traditional manufacturing sectors: A review,”
XAI [108]. Some notable benefits are reduced latency Computers in Industry, vol. 125, p. 103390, 2021.
through faster computation at the edge, anomaly detec- [5] A. Palange and P. Dhatrak, “Lean manufacturing a vital tool to en-
tion and energy efficiency. Though these technologies hance productivity in manufacturing,” Materials Today: Proceedings,
vol. 46, pp. 729–736, 2021.
are individually beneficial, integration would result [6] H. Zhang, H. Xu, X. Tian, J. Jiang, and J. Ma, “Image fusion
in enormous challenges like scalability and lack of meets deep learning: A survey and perspective,” Information Fusion,
standardization. vol. 76, pp. 323–336, 2021.
[7] R. Gutiérrez, V. Rampérez, H. Paggi, J. A. Lara, and J. Soriano,
“On the use of information fusion techniques to improve informa-
tion quality: Taxonomy, opportunities and challenges,” Information
Fusion, vol. 78, pp. 102–137, 2022.
V. Conclusion [8] A. Y. B. Ahmad, T. N. Gongada, G. Shrivastava, R. S. Gabbi, S. Is-
AI has faced adoption challenges due to its lack of trans- lam, and K. Nagaraju, “E-commerce trend analysis and management
for industry 5.0 using user data analysis,” International Journal of
parency and reliance on diverse data. Also, information Intelligent Systems and Applications in Engineering, vol. 11, no. 11s,
fusion is vital for AI-based applications as it learns through pp. 135–150, 2023.
the assorted data collected from varied sources. The trust [9] A. Bécue, I. Praça, and J. Gama, “Artificial intelligence, cyber-threats
and industry 4.0: Challenges and opportunities,” Artificial Intelligence
in AI systems was significantly regained with the evolution Review, vol. 54, no. 5, pp. 3849–3886, 2021.
of XAI with its explanations for AI’s decisions. Information [10] M. A. Alohali, F. N. Al-Wesabi, A. M. Hilal, S. Goel, D. Gupta,
fusion is required for XAI to derive optimal decisions and and A. Khanna, “Artificial intelligence enabled intrusion detection
systems for cognitive cyber-physical systems in industry 4.0 envi-
appropriate explanations to intended users. As I5.0 and its ronment,” Cognitive Neurodynamics, vol. 16, no. 5, pp. 1045–1057,
applications rely on AI, XAI for I5.0 would significantly 2022.
enhance the performance of the Industrial Revolution and its [11] S. Huang, B. Wang, X. Li, P. Zheng, D. Mourtzis, and L. Wang,
“Industry 5.0 and society 5.0—comparison, complementation and co-
trust in adoption. Therefore, this survey presents the concept evolution,” Journal of manufacturing systems, vol. 64, pp. 424–428,
of utilizing XAI for I5.0, the need for information fusion, 2022.
viable applications, and challenges in adoption and future [12] S. Rani, G. Srivastava et al., “Secure hierarchical fog computing-
based architecture for industry 5.0 using an attribute-based encryption
research directions. The survey first provides insights into scheme,” Expert Systems with Applications, vol. 235, p. 121180,
the need for XAI for I5.0 and briefs the major contributions 2024.
to the research. Followed by the historical background of the [13] X. Xu, Y. Lu, B. Vogel-Heuser, and L. Wang, “Industry 4.0 and
industry 5.0—inception, conception and perception,” Journal of Man-
Industrial Revolution, the taxonomy of XAI, its framework, ufacturing Systems, vol. 61, pp. 530–535, 2021.
and the motivation behind the amalgamation. Consequently, [14] J. Leng, W. Sha, B. Wang, P. Zheng, C. Zhuang, Q. Liu, T. Wuest,
metaverse applications in different verticals of I5.0, namely D. Mourtzis, and L. Wang, “Industry 5.0: Prospect and retrospect,”
Journal of Manufacturing Systems, vol. 65, pp. 279–295, 2022.
Smart Factories, Healthcare, Governance, Transportation, [15] P. K. R. Maddikunta, Q.-V. Pham, B. Prabadevi, N. Deepa, K. Dev,
Education 5.0, Agriculture 5.0, and Energy 5.0, have various T. R. Gadekallu, R. Ruby, and M. Liyanage, “Industry 5.0: A survey
implementation issues. Finally, the open issues and chal- on enabling technologies and potential applications,” Journal of
Industrial Information Integration, vol. 26, p. 100257, 2022.
lenges, namely, Transparency, interoperability, and security, [16] A. Akundi, D. Euresti, S. Luna, W. Ankobiah, A. Lopes, and
were identified, with feasible solutions for resolving those I. Edinbarough, “State of industry 5.0—analysis and identification
issues. Furthermore, insights for future research directions of current research trends,” Applied System Innovation, vol. 5, no. 1,
p. 27, 2022.
were also discussed.
[17] C. R. Morris, The dawn of innovation: The first American industrial
Future research would focus on developing an XAI frame- revolution. PublicAffairs, 2012.
work for providing user-specific explanations in different [18] A. E. Musson and E. Robinson, Science and technology in the
application scenarios of I5.0 by utilizing Human-Computer industrial revolution. Taylor & Francis, 1989, vol. 3.
[19] Y. Lu, “Industry 4.0: A survey on technologies, applications and open
Interaction and Natural Language Processing tools for multi- research issues,” Journal of industrial information integration, vol. 6,
lingual different user-level explanations. Furthermore, the pp. 1–10, 2017.
safety of these systems can be enhanced by adopting fed- [20] M. Breque, L. De Nul, and A. Petridis, “Industry 5.0: towards a
sustainable, human-centric and resilient european industry,” Luxem-
erated learning and can be efficient and fastened with less bourg, LU: European Commission, Directorate-General for Research
resource utilization through transfer learning models. and Innovation, 2021.

VOLUME , 21

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

TABLE 3. Open Issues, Challenges, and Future Directions for XAI in Industry 5.0

Open Issue Description Challenges Future Directions Impact Importance

Explainability-by-Design Incorporating explainability principles Balancing complexity and transparency, Develop frameworks and tools that in- Enhances trust and user High
from the beginning of the AI model ensuring explainability without sacrific- tegrate explainability at all stages of AI understanding of AI sys-
development to enhance transparency ing model performance. development. tems.
and performance.
Context-aware Developing explanations that are spe- Adapting explanations to various in- Design adaptive systems that tailor ex- Increases effectiveness of High
Explanations cific to industry sectors to improve dustrial contexts and user roles while planations based on the sector-specific AI in specialized sectors.
human-machine collaboration and oper- maintaining accuracy and relevance. needs and user roles.
ational objectives.
Trustworthy Explanations Enhancing the trust, simplicity, and ac- Developing methods to effectively sim- Focus on methods that provide deeper Critical for adopting AI in Very High
for Complex Models curacy of deep learning models through plify complex model outputs into un- insights into complex model behaviors sensitive applications.
XAI, making their operations less derstandable explanations. and decision-making processes.
opaque.
Real-time Explainability Designing XAI methods that can de- Creating fast and efficient explanation Develop lightweight, efficient algo- Essential for time- Very High
liver explanations in real-time to align processes that do not compromise the rithms capable of operating under strict sensitive operations like
with the quick decision-making needed timeliness of decision-making. time constraints. manufacturing.
in industrial applications.
Integration with Diverse Creating specialized XAI methods that Managing and synthesizing information Innovate cross-modal explanation tech- Supports better decision- High
Data Formats can handle and explain multimodal data from heterogeneous data sources to pro- niques that seamlessly integrate differ- making from complex
from various industrial sources. vide coherent and actionable explana- ent data types. data sources.
tions.
Human-centered Interac- Improving interfaces and trust through Designing intuitive interfaces that facil- Enhance interface designs and develop Improves usability and High
tion and Trust XAI explanations, ensuring that human- itate understanding and trust between trust-building features within XAI. acceptance of automated
machine collaboration is effective and users and AI systems. systems.
reliable.
Standardization and Eval- Developing standards and metrics for Establishing universally accepted met- Establish international standards for Facilitates consistency and Very High
uation evaluating and implementing XAI in rics and standards in a rapidly evolving XAI applications and evaluation comparability across sys-
I5.0 to guide technology usage and val- field. methodologies. tems.
idation.
Ethical Considerations Ensuring XAI methods comply with Navigating varying international data Create ethical guidelines and compli- Protects user data and ad- Very High
and Regulations ethical concerns and data privacy regu- privacy laws and ethical standards. ance checks within XAI systems. heres to legal standards.
lations to maintain security and privacy.
Transparency Addressing the need for comprehensive Overcoming the complexity of AI mod- Advance methods that offer more de- Builds trust and enables High
transparency in XAI systems to foster els to provide understandable and use- tailed insights into AI operations and better oversight of AI sys-
user confidence and comprehension of ful explanations. decision-making. tems.
AI decisions.
Interpretability Ensuring XAI models are interpretable Achieving high levels of interpretability Focus on developing techniques that Ensures AI decisions are Very High
so users can understand how decisions without impacting the performance of offer clear, understandable explanations justifiable and account-
are made, which is crucial for trust and complex AI systems. suited to non-expert users. able.
regulatory compliance.
Security Focusing on securing XAI solutions to Protecting against adversarial attacks Strengthen defenses against attacks on Enhances the safety and Very High
protect sensitive data and prevent AI and ensuring the integrity of explana- AI models and develop secure protocols reliability of AI applica-
system abuse, addressing threats like tions in the presence of potential secu- for explanation generation. tions.
adversarial attacks and ensuring robust rity vulnerabilities.
explanations.

[21] M. Javaid, A. Haleem, R. P. Singh, M. I. U. Haq, A. Raina, and [27] A. Hassan, P. K. Dutta, S. Gupta, E. Mattar, and S. Singh, Human-
R. Suman, “Industry 5.0: Potential applications in covid-19,” Journal Centered Approaches in Industry 5.0: Human-Machine Interaction,
of Industrial Integration and Management, vol. 5, no. 04, pp. 507– Virtual Reality Training, and Customer Sentiment Analysis: Human-
530, 2020. Machine Interaction, Virtual Reality Training, and Customer Senti-
[22] Z. Liao, L. Tu, X. Li, X.-J. Liang, and S. Huo, “Virus-inspired ment Analysis. IGI Global, 2024.
nanosystems for drug delivery,” Nanoscale, vol. 13, no. 45, pp. [28] H. Alimam, G. Mazzuto, N. Tozzi, F. E. Ciarapica, and M. Bevilac-
18 912–18 924, 2021. qua, “The resurrection of digital triplet: A cognitive pillar of human-
[23] E. Roels, S. Terryn, J. Brancart, F. Sahraeeazartamar, F. Clemens, machine integration at the dawn of industry 5.0,” Journal of King
G. Van Assche, and B. Vanderborght, “Self-healing sensorized soft Saud University-Computer and Information Sciences, p. 101846,
robots,” Materials Today Electronics, vol. 1, p. 100003, 2022. 2023.
[29] U. A. Khan, N. M. Khan, and M. H. Zafar, “Resource efficient pv
[24] R. Kumar, L. Sharma, R. Chhibber, A. Dixit, and R. Singhal, power forecasting: Transductive transfer learning based hybrid deep
“Environmental degradation of glass fiber-reinforced nanocomposites learning model for smart grid in industry 5.0,” Energy Conversion
with self-healing reinforcement in polymer matrix for wind turbine and Management: X, vol. 20, p. 100486, 2023.
blade application,” Transactions of the Indian Institute of Metals, [30] B. Bajic, N. Suzic, S. Moraca, M. Stefanović, M. Jovicic, and
vol. 74, no. 12, pp. 3119–3133, 2021. A. Rikalovic, “Edge computing data optimization for smart quality
[25] K. Fang, J. Qiu, T. Wang, K. Zheng, L. Xing, K. Mao, and K. Chi, management: Industry 5.0 perspective,” Sustainability, vol. 15, no. 7,
“Idres: Identity-based respiration monitoring system for digital twins p. 6032, 2023.
enabled healthcare,” IEEE Journal on Selected Areas in Communi- [31] S. Wang, M. A. Qureshi, L. Miralles-Pechuán, T. Huynh-The, T. R.
cations, 2023. Gadekallu, and M. Liyanage, “Explainable ai for 6g use cases:
[26] B. Wang, H. Zhou, X. Li, G. Yang, P. Zheng, C. Song, Y. Yuan, Technical aspects and research challenges,” IEEE Open Journal of
T. Wuest, H. Yang, and L. Wang, “Human digital twin in the context the Communications Society, 2024.
of industry 5.0,” Robotics and Computer-Integrated Manufacturing, [32] D. Gunning, “Explainable artificial intelligence (xai),” Defense ad-
vol. 85, p. 102626, 2024. vanced research projects agency (DARPA), nd Web, vol. 2, no. 2,

22 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

p. 1, 2017. [52] A. Rahman, M. Hossain, G. Muhammad, D. Kundu, T. Debnath,

[33] P. J. Phillips, C. A. Hahn, P. C. Fontana, D. A. Broniatowski, and M. Rahman, M. Khan, S. Islam, P. Tiwari, S. S. Band et al., “Fed-
M. A. Przybocki, “Four principles of explainable artificial intelli- erated learning-based ai approaches in smart healthcare: concepts,
gence,” Gaithersburg, Maryland, p. 18, 2020. taxonomies, challenges and open issues,” Cluster Computing, pp. 1–
[34] F. Xu, H. Uszkoreit, Y. Du, W. Fan, D. Zhao, and J. Zhu, “Explainable 41, 2022.
ai: A brief survey on history, research areas, approaches and chal- [53] D. Dave, H. Naik, S. Singhal, and P. Patel, “Explainable ai
lenges,” in Natural Language Processing and Chinese Computing: meets healthcare: A study on heart disease dataset,” arXiv preprint
8th CCF International Conference, NLPCC 2019, Dunhuang, China, arXiv:2011.03195, 2020.
October 9–14, 2019, Proceedings, Part II 8. Springer, 2019, pp. [54] D. Saraswat, P. Bhattacharya, A. Verma, V. K. Prasad, S. Tanwar,
563–574. G. Sharma, P. N. Bokoro, and R. Sharma, “Explainable ai for
[35] T. Speith, “A review of taxonomies of explainable artificial in- healthcare 5.0: opportunities and challenges,” IEEE Access, 2022.
telligence (xai) methods,” in 2022 ACM Conference on Fairness, [55] A. Adel, “Future of industry 5.0 in society: human-centric solutions,
Accountability, and Transparency, 2022, pp. 2239–2250. challenges and prospective research areas,” Journal of Cloud Com-
[36] J. Dieber and S. Kirrane, “Why model why? assessing the strengths puting, vol. 11, no. 1, p. 40, 2022.
and limitations of lime,” arXiv preprint arXiv:2012.00093, 2020. [56] D. Leslie, “Understanding artificial intelligence ethics and safety,”
[37] S. M. Lundberg and S.-I. Lee, “A unified approach arXiv preprint arXiv:1906.05684, 2019.
to interpreting model predictions,” in Advances in Neural [57] A. Reñones, D. Dalle Carbonare, and S. Gusmeroli, “European big
Information Processing Systems 30, I. Guyon, U. V. Luxburg, data value association position paper on the smart manufacturing
S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and industry,” Enterprise Interoperability: Smart Services and Business
R. Garnett, Eds. Curran Associates, Inc., 2017, pp. 4765– Impact of Enterprise Interoperability, pp. 179–185, 2018.
4774. [Online]. Available: http://papers.nips.cc/paper/7062-a-unified- [58] J. M. Rožanec, I. Novalija, P. Zajec, K. Kenda, H. Tavakoli, S. Suh,
approach-to-interpreting-model-predictions.pdf E. Veliou, D. Papamartzivanos, T. Giannetsos, S. A. Menesidou et al.,
[38] M. Chromik, “reshape: A framework for interactive explanations in “Human-centric artificial intelligence architecture for industry 5.0
xai based on shap,” in Proceedings of 18th European Conference applications,” arXiv preprint arXiv:2203.10794, 2022.
on Computer-Supported Cooperative Work. European Society for [59] F. Sovrano, F. Vitali, and M. Palmirani, “Making things explainable
Socially Embedded Technologies (EUSSET), 2020. vs explaining: Requirements and challenges under the gdpr,” in
[39] K. Zhang, P. Xu, and J. Zhang, “Explainable ai in deep reinforcement International Workshop on AI Approaches to the Complexity of
learning models: A shap method applied in power system emergency Legal Systems, International Workshop on AI Approaches to the
control,” in 2020 IEEE 4th Conference on Energy Internet and Energy Complexity of Legal Systems, International Workshop on Explainable
System Integration (EI2). IEEE, 2020, pp. 711–716. and Responsible AI and Law. Springer, 2021, pp. 169–182.
[40] R. Younisse, A. Ahmad, and Q. Abu Al-Haija, “Explaining intrusion [60] A. Chandel and B. Sharma, “Technology aspects of artificial intel-
detection-based convolutional neural networks using shapley additive ligence: Industry 5.0 for organization decision making,” in Interna-
explanations (shap),” Big Data and Cognitive Computing, vol. 6, tional Conference on Information Systems and Management Science.
no. 4, p. 126, 2022. Springer, 2023, pp. 79–90.
[41] K. Roshan and A. Zafar, “Utilizing xai technique to improve [61] U. Wajid, A. Nizamis, and V. Anaya, “Towards industry 5.0–a
autoencoder based model for computer network anomaly detec- trustworthy ai framework for digital manufacturing with humans in
tion with shapley additive explanation (shap),” arXiv preprint control,” Proceedings http://ceur-ws. org ISSN, vol. 1613, p. 0073,
arXiv:2112.08442, 2021. 2022.
[42] S. Wang, M. Atif Qureshi, L. Miralles-Pechuán, T. Reddy Gadekallu, [62] I. Sharma, I. Garg, and D. Kiran, “Industry 5.0 and smart cities:
M. Liyanage et al., “Explainable ai for b5g/6g: technical aspects, use A futuristic approach,” European Journal of Molecular & Clinical
cases, and research challenges,” arXiv e-prints, pp. arXiv–2112, 2021. Medicine, vol. 7, no. 08, pp. 2515–8260, 2020.
[43] E. Costa, “Industry 5.0 and sdg 9: a symbiotic dance towards [63] K. Fang, T. Wang, L. Tong, X. Fang, Y. Pan, W. Wang, and J. Li,
sustainable transformation,” Sustainable Earth Reviews, vol. 7, no. 1, “Non-intrusive security assessment methods for future autonomous
p. 4, 2024. transportation iov,” IEEE Transactions on Automation Science and
[44] V. Özdemir and N. Hekim, “Birth of industry 5.0: Making sense Engineering, 2023.
of big data with artificial intelligence,“the internet of things” and [64] Y. Zhang, G. Zhang, R. Fierro, and Y. Yang, “Force-driven traffic
next-generation technology policy,” Omics: a journal of integrative simulation for a future connected autonomous vehicle-enabled smart
biology, vol. 22, no. 1, pp. 65–76, 2018. transportation system,” IEEE Transactions on Intelligent Transporta-
[45] S. Wang, J. Wan, D. Zhang, D. Li, and C. Zhang, “Towards smart tion Systems, vol. 19, no. 7, pp. 2221–2233, 2018.
factory for industry 4.0: a self-organized multi-agent system with big [65] S. Latif, H. Afzaal, and N. A. Zafar, “Intelligent traffic monitoring
data based feedback and coordination,” Computer networks, vol. 101, and guidance system for smart city,” in 2018 International Con-
pp. 158–168, 2016. ference on Computing, Mathematics and Engineering Technologies
[46] L. Xinde, F. DUNKIN, and J. DEZERT, “Multi-source information (iCoMET). IEEE, 2018, pp. 1–6.
fusion: Progress and future,” Chinese Journal of Aeronautics, 2023. [66] E. Fantin Irudaya Raj and M. Appadurai, “Internet of things-based
[47] G. Dhiman, A. Nagar, and S. Kadry, “Explainable artificial intelli- smart transportation system for smart cities,” in Intelligent systems
gence for the social internet of things: Analysis and modeling using for social good. Springer, 2022, pp. 39–50.
collaborative technologies [special section editorial],” IEEE Systems, [67] T. Kuncara, A. S. Putra, N. Aisyah, and V. Valentino, “Effectiveness
Man, and Cybernetics Magazine, vol. 8, no. 4, pp. 34–35, 2022. of the e-ticket system using qr codes for smart transportation sys-
[48] O. Serradilla, E. Zugasti, J. Ramirez de Okariz, J. Rodriguez, and tems,” International Journal of Science, Technology & Management,
U. Zurutuza, “Adaptable and explainable predictive maintenance: vol. 2, no. 3, pp. 900–907, 2021.
Semi-supervised deep learning for anomaly detection and diagnosis [68] D. Bamwesigye and P. Hlavackova, “Analysis of sustainable transport
in press machine data,” Applied Sciences, vol. 11, no. 16, p. 7376, for smart cities,” Sustainability, vol. 11, no. 7, p. 2140, 2019.
2021. [69] M. Lahby, U. Kose, and A. K. Bhoi, Explainable Artificial Intelli-
[49] C. Park, S. Lim, D. Cha, and J. Jeong, “Fv-ad: F-anogan based gence for Smart Cities. CRC Press, 2021.
anomaly detection in chromate process for smart manufacturing,” [70] S. Atakishiyev, M. Salameh, H. Yao, and R. Goebel, “Explain-
Applied Sciences, vol. 12, no. 15, p. 7549, 2022. able artificial intelligence for autonomous driving: A comprehen-
[50] F. Jiang, Y. Jiang, H. Zhi, Y. Dong, H. Li, S. Ma, Y. Wang, Q. Dong, sive overview and field guide for future research directions,” arXiv
H. Shen, and Y. Wang, “Artificial intelligence in healthcare: past, preprint arXiv:2112.11561, 2021.
present and future,” Stroke and vascular neurology, vol. 2, no. 4, [71] A. Nascita, A. Montieri, G. Aceto, D. Ciuonzo, V. Persico, and
2017. A. Pescapé, “Xai meets mobile traffic classification: Understanding
[51] M. Ghassemi, L. Oakden-Rayner, and A. L. Beam, “The false hope and improving multimodal deep learning architectures,” IEEE Trans-
of current approaches to explainable artificial intelligence in health actions on Network and Service Management, vol. 18, no. 4, pp.
care,” The Lancet Digital Health, vol. 3, no. 11, pp. e745–e750, 2021. 4225–4246, 2021.

VOLUME , 23

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Open Journal of the Communications Society. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJCOMS.2024.3473891

[72] L. Gaur and B. M. Sahoo, “Introduction to explainable ai and [94] M. Kuzlu, U. Cali, V. Sharma, and Ö. Güler, “Gaining insight into
intelligent transportation,” in Explainable Artificial Intelligence for solar photovoltaic power generation forecasting utilizing explainable
Intelligent Transportation Systems. Springer, 2022, pp. 1–25. artificial intelligence tools,” IEEE Access, vol. 8, pp. 187 814–
[73] D. Zar, N. Hazon, and A. Azaria, “Explaining ridesharing: selection 187 823, 2020.
of explanations for increasing user satisfaction,” in European Con- [95] A. Yayla, L. Haghnegahdar, and E. Dincelli, “Explainable artificial
ference on Multi-Agent Systems. Springer, 2021, pp. 89–107. intelligence for smart grid intrusion detection systems,” IT Profes-
[74] M. H. Kabir, K. F. Hasan, M. K. Hasan, and K. Ansari, “Explainable sional, vol. 24, no. 5, pp. 18–24, 2022.
artificial intelligence for smart city application: A secure and trusted [96] L. Coroama and A. Groza, “Evaluation metrics in explainable ar-
platform,” in Explainable Artificial Intelligence for Cyber Security. tificial intelligence (xai),” in International conference on advanced
Springer, 2022, pp. 241–263. research in technologies, information, innovation and sustainability.
[75] E. G. Carayannis and J. Morawska-Jancelewicz, “The futures of Springer, 2022, pp. 401–413.
europe: Society 5.0 and industry 5.0 as driving forces of future [97] N. Balasubramaniam, M. Kauppinen, A. Rannisto, K. Hiekkanen,
universities,” Journal of the Knowledge Economy, pp. 1–27, 2022. and S. Kujala, “Transparency and explainability of ai systems:
[76] J. M. Sari and E. Purwanta, “The implementation of artificial intelli- From ethical guidelines to requirements,” Information and Software
gence in stem-based creative learning in the society 5.0 era,” Tadris: Technology, vol. 159, p. 107197, 2023.
Jurnal Keguruan dan Ilmu Tarbiyah, vol. 6, no. 2, pp. 433–440, 2021. [98] U. Ehsan, Q. V. Liao, M. Muller, M. O. Riedl, and J. D. Weisz, “Ex-
[77] D. McArthur, M. Lewis, and M. Bishary, “The roles of artificial panding explainability: Towards social transparency in ai systems,”
intelligence in education: current progress and future prospects,” in Proceedings of the 2021 CHI Conference on Human Factors in
Journal of Educational Technology, vol. 1, no. 4, pp. 42–80, 2005. Computing Systems, 2021, pp. 1–19.
[78] G. Ramaswami, T. Susnjak, and A. Mathrani, “Supporting students’ [99] A. A. Tubella, A. Theodorou, V. Dignum, and F. Dignum, “Gover-
academic performance using explainable machine learning with au- nance by glass-box: Implementing transparent moral bounds for ai
tomated prescriptive analytics,” Big Data and Cognitive Computing, behaviour,” arXiv preprint arXiv:1905.04994, 2019.
vol. 6, no. 4, p. 105, 2022. [100] P. D. Winter and A. Carusi, “(de) troubling transparency: artificial
[79] C. Conati, O. Barral, V. Putnam, and L. Rieger, “Toward person- intelligence (ai) for clinical applications,” Medical Humanities, 2022.
alized xai: A case study in intelligent tutoring systems,” Artificial [101] D. Shin, “The effects of explainability and causability on perception,
Intelligence, vol. 298, p. 103503, 2021. trust, and acceptance: Implications for explainable ai,” International
[80] P. Hur, H. LEE, S. Bhat, and N. Bosch, “Using machine learning Journal of Human-Computer Studies, vol. 146, p. 102551, 2021.
explainability methods to personalize interventions for students,” in [102] M. Nourani, C. Roy, D. R. Honeycutt, E. D. Ragan, and V. Gogate,
Proceedings of the 15th International Conference on Educational “Detoxer: A visual debugging tool with multiscope explanations for
Data Mining, A. Mitrovic and N. Bosch, Eds. International Educa- temporal multilabel classification,” IEEE Computer Graphics and
tional Data Mining Society, 2022, pp. 438–445. Applications, vol. 42, no. 6, pp. 37–46, 2022.
[81] V. Saiz-Rubio and F. Rovira-Más, “From smart farming towards [103] A. Theodorou and V. Dignum, “Towards ethical and socio-legal
agriculture 5.0: A review on crop data management,” Agronomy, governance in ai,” Nature Machine Intelligence, vol. 2, no. 1, pp.
vol. 10, no. 2, p. 207, 2020. 10–12, 2020.
[82] D. R. Vincent, N. Deepa, D. Elavarasan, K. Srinivasan, S. H. [104] A. Kuppa and N.-A. Le-Khac, “Black box attacks on explainable
Chauhdary, and C. Iwendi, “Sensors driven ai-based agriculture rec- artificial intelligence (xai) methods in cyber security,” in 2020 In-
ommendation model for assessing land suitability,” Sensors, vol. 19, ternational Joint Conference on Neural Networks (IJCNN). IEEE,
no. 17, p. 3667, 2019. 2020, pp. 1–8.
[83] N. Deepa and K. Ganesan, “Hybrid rough fuzzy soft classifier based [105] ——, “Adversarial xai methods in cybersecurity,” IEEE transactions
multi-class classification model for agriculture crop selection,” Soft on information forensics and security, vol. 16, pp. 4924–4938, 2021.
computing, vol. 23, no. 21, pp. 10 793–10 809, 2019. [106] T. N. Nguyen and R. Choo, “Human-in-the-loop xai-enabled vul-
[84] ——, “Mahalanobis taguchi system based criteria selection tool for nerability detection, investigation, and mitigation,” in 2021 36th
agriculture crops,” Sādhanā, vol. 41, no. 12, pp. 1407–1414, 2016. IEEE/ACM International Conference on Automated Software Engi-
[85] N. Deepa, K. Srinivasan, C.-Y. Chang, and A. K. Bashir, “An efficient neering (ASE). IEEE, 2021, pp. 1210–1212.
ensemble vtopes multi-criteria decision-making model for sustainable [107] P. Steinmüller, T. Schulz, F. Graf, and D. Herr, “explainable ai for
sugarcane farms,” Sustainability, vol. 11, no. 16, p. 4288, 2019. quantum machine learning,” arXiv preprint arXiv:2211.01441, 2022.
[86] M. Ryo, “Explainable artificial intelligence and interpretable machine [108] A. Vasiliev and I. Brusakova, “Innovative design of marketing ecosys-
learning for agricultural data analysis,” Artificial Intelligence in tems,” MANAGEMENT SCIENCES, vol. 14, no. 1, p. 88, 2024.
Agriculture, vol. 6, pp. 257–265, 2022.
[87] A. Lad and M. S. Raval, “Resolving issues with wheat head detection:
A use case of xai in agriculture scenario,” Lecture Notes in Computer
Science, pp. 1–6, 2021.
[88] Q. H. Ngo, T. Kechadi, and N.-A. Le-Khac, “Oak4xai: Model towards
out-of-box explainable artificial intelligence for digital agriculture,” in
International Conference on Innovative Techniques and Applications
of Artificial Intelligence. Springer, 2022, pp. 238–251.
[89] S. Kinger and V. Kulkarni, “Explainable ai for deep learning based
disease detection,” in 2021 Thirteenth International Conference on
Contemporary Computing (IC3-2021), 2021, pp. 209–216.
[90] J. Deng, F. Wang, Y. Chen, and X. Zhao, “From industries 4.0 to
energy 5.0: concept and framework of intelligent energy systems,”
Acta Automatica Sinica, vol. 41, no. 12, pp. 2003–2016, 2015.
[91] R. Machlev, L. Heistrene, M. Perl, K. Levy, J. Belikov, S. Mannor,
and Y. Levron, “Explainable artificial intelligence (xai) techniques for
energy and power systems: Review, challenges and opportunities,”
Energy and AI, p. 100169, 2022.
[92] C. Turner, O. Okorie, and J. Oyekan, “Xai sustainable human in the
loop maintenance,” IFAC-PapersOnLine, vol. 55, no. 19, pp. 67–72,
2022.
[93] E. Henriksen, U. Halden, M. Kuzlu, and U. Cali, “Electrical load
forecasting utilizing an explainable artificial intelligence (xai) tool on
norwegian residential buildings,” in 2022 International Conference on
Smart Energy Systems and Technologies (SEST). IEEE, 2022, pp.
1–6.

24 VOLUME ,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/