The 12th Player: Explainable Artiﬁcial Intelligence (XAI) in Football:

Conceptualisation, Applications, Challenges and Future Directions

Andria Procopiou and Andriani Piki

School of Sciences, University of Central Lancashire Cyprus, Larnaca, Cyprus

Keywords:

Explainable Artiﬁcial Intelligence, Machine Learning, Deep Learning, Football Analytics, Injury Prediction,

Rehabilitation, Football Tactical Analysis, Human Factors, Human-Centred AI.

Abstract:

Artiﬁcial intelligence (AI) has demonstrated tremendous progress in many domains, especially with the vast

deployment of machine and deep learning. Recently, AI has been introduced to the sports domain including

the football (soccer) industry with applications in injury prediction and tactical analysis. However, the fact

remains that the more complex an AI model is, the less explainable it becomes. Its black-box nature makes it

difﬁcult for human operators to understand its results, interpret its decisions and ultimately trust the model it-

self. This problem is magniﬁed when the decisions and results suggested by an AI model affect the functioning

of complex and multi-layered systems and entities, with a football club being such an example. Explainable

artiﬁcial intelligence (XAI) has emerged for making an AI model more explainable, understandable and in-

terpretable, thus assisting the creation of human-centered AI models. This paper discusses how XAI could be

applied in the football domain to beneﬁt both the players and the club.

1 INTRODUCTION

During the last decade the advances in artiﬁcial in-

telligence (AI) have been signiﬁcant to the world.

Machine and deep learning (perhaps the most popu-

lar subsets of AI) have been integrated in numerous

technology-enabled sectors (Russell, 2010) forming a

vital part in their operations. Beyond the corporate

sector, AI became a vital part in people’s lives, im-

proving their everyday tasks and quality of life. Ex-

amples include voice recognition, self-driving cars,

and recommendation systems (Gupta et al., 2021),

(Bharati et al., 2020a), (Mondal et al., 2021), (Bharati

et al., 2020b). AI is also being deployed in the sports

industry including football (Moustakidis et al., 2023).

The main goal of utilising AI is to improve the de-

cision making in numerous areas. Notable exam-

ples include improving individual players’ and the

team’s performance and development (Moustakidis

et al., 2023), assisting in scouting, identifying poten-

tial talents from the youth academies, enhancing ana-

lytics and tactics, predicting and potentially prevent-

ing injuries and optimising rehabilitation periods for

injured players (Rathi et al., 2020).

The sophisticated learning, reasoning, and adap-

tation capabilities of AI models have proved to be

pivotal in all of the areas it has been utilised, includ-

ing the sports industry, and require little or no hu-

man intervention (Arrieta et al., 2020). However, the

fact that machine and deep learning algorithms follow

a black-box approach (Arrieta et al., 2020) presents

challenges in explaining a model’s decisions and pre-

dictions. AI models do not provide sufﬁcient justiﬁca-

tion, explainability, and interpretability on their over-

all behaviour (Gunning et al., 2019). Evidently, clear

and precise explanations are vital towards demys-

tifying how AI operates, especially when deployed

in complex and multi-layered domains such as foot-

ball, where the decisions and predictions may affect

a team’s performance and even more importantly a

player’s health.

Gaining an in-depth understanding of how AI sys-

tems function and achieve their results is needed,

especially when critical decisions need to be made

(Goodman and Flaxman, 2017). In the sports indus-

try, especially football, decisions are made on a daily

basis with regards to multiple operations of a club, in-

cluding scouting procedures, team tactics, opposition

analysis, players’ personal ﬁtness/rehabilitation pro-

grammes and injury prevention procedures. Even a

small miscalculation can be proven ﬁnancially costly

for the club. In addition, the players’ physical and

mental health and overall wellbeing can be negatively

impacted. Therefore, there is need for improved ex-

Procopiou, A. and Piki, A.

The 12th Player: Explainable Artiﬁcial Intelligence (XAI) in Football: Conceptualisation, Applications, Challenges and Future Directions.

DOI: 10.5220/0012233800003587

In Proceedings of the 11th International Conference on Sport Sciences Research and Technology Support (icSPORTS 2023), pages 213-220

ISBN: 978-989-758-673-6; ISSN: 2184-3201

213

plainability and interpreatability of AI models. Fur-

thermore, there is need for human experts to reas-

sure both themselves and relevant stakeholders that

AI solutions are transparent, unbiased, and trustwor-

thy, by essentially minimising its black-box nature

(Miller, 2019). Explainable Artiﬁcial Intelligence

(XAI) emerged to make AI systems’ reasoning, out-

puts and overall results more understandable and clear

to human experts (Chazette et al., 2021), (K

ohl et al.,

2019), (Langer et al., 2021b), (P

aez, 2019).

Motivated by the eminent concerns and opportuni-

ties observed, we discuss how XAI could be utilised

in the football industry, focusing in various operations

impacting both the team and individual stakehold-

ers (players, coaches, sports analysts, etc.). Firstly,

we provide the necessary background knowledge dis-

cussing key concepts of XAI, including deﬁnitions,

the characteristics an XAI model should exhibit, and

how XAI could become more human-centred. Sub-

sequently, we discuss in-depth how XAI could be

utilised in various aspects of the football industry,

including scouting, tactical analysis, player develop-

ment and performance optimisation, injury prediction

and prevention and ﬁnally, rehabilitation. We con-

clude by summarising our insights and provide future

directions on XAI applied in the football industry.

2 RESEARCH BACKGROUND

2.1 XAI Deﬁnitions

Numerous deﬁnitions have been proposed to capture

the meaning of XAI. The most notable ones deﬁne

XAI, in general terms, as one of AI’s sub-ﬁelds that

is responsible for accompanying an AI model with in-

telligible explanations to the end users and stakehold-

ers by constructing effective and accurate approaches

(Van Lent et al., 2004) (Biran and Cotton, 2017),

(Miller, 2019), (Mittelstadt et al., 2019). More specif-

ically, XAI has been deﬁned as the set of features

that assist users understanding how an AI model con-

structs its predictions (Arrieta et al., 2020), (Nazar

et al., 2021). To this end, XAI should highlight the

most inﬂuential factors contributing to the prediction

process (Viton et al., 2020), hence allowing more ef-

ﬁcient pre-processing to be conducted.

Making an AI model more understandable and ex-

plainable does not directly guarantee its interpreata-

bility, especially when it is used by multiple types of

users with different levels of expertise and knowledge

and varying degrees of experience. This issue was

correctly raised by (Bharati et al., 2023). Speciﬁcally,

it is argued that explainability focuses on the ’why’

behind a decision and not on the ’how’, while in-

tepretability deals with making the users understand

the rationale behind its decision (Vishwarupe et al.,

2022), (Miller, 2019).

An important factor to consider when it comes

to XAI is the need for reassuring the human ex-

perts/users that the AI model constructed is neither

biased nor discriminating (Duell et al., 2021). XAI

plays a pivotal role in strengthening trust between hu-

man experts and AI models and assisting in better

collaboration between the two (Adadi and Berrada,

2018). In conclusion, the authors in (Gunning et al.,

2019) effectively stated that XAI should be heavily in-

ﬂuenced by the social sciences (Miller, 2019), specif-

ically the psychology of explanation, and should aim

to make machine and deep learning algorithms:

• More explainable and socially aware, while also

maintaining their high levels of accuracy.

• Human-centred, by assisting in creating and

maintaining trust between human experts and AI.

2.2 XAI Characteristics and Goals

According to (Arrieta et al., 2020), the main proper-

ties an XAI model include:

• Understandability (also called Intelligibility):

An AI model should make its functioning and

decision-making more understandable to humans.

Low-level details such as its inner structure and

algorithmic training procedure should be omitted.

• Comprehensibility: An AI model should provide

human-readable explanations regarding its learnt

knowledge. In that way, its complexity becomes

more approachable to human operators.

• Interpretability: An AI model should describe re-

sults in a human-meaningful way.

• Explainability: An AI model should provide

meaningful explanations about the results to assist

human experts in their decisions.

• Transparency: An AI model should be fair, unbi-

ased, and transparent without leaving any details

or knowledge hidden, vague or ambiguous to hu-

man experts.

2.3 User-Centred XAI

The most important objective of XAI is to pro-

vide substantial support to the various types of users

in accurate, effective and correct decision-making

(Nadeem et al., 2022), thus demystifying AI’s black-

box nature. We proceed with discussing a set of com-

mon practices with which XAI can be realised in prac-

tice based on previous work by (Nadeem et al., 2022).

icSPORTS 2023 - 11th International Conference on Sport Sciences Research and Technology Support

214

2.3.1 XAI Enabled by Visualisation

The construction of visualisations is the most popu-

lar and straightforward way of providing explainabil-

ity and understandability of how an AI model oper-

ates. This is due to human cognition being in favour

of visual information instead of text when it comes

to decision making (Padilla et al., 2018). There are

multiple parts of an AI model that can be visualised.

One approach is to visualise how an AI model pro-

ceeds to make a decision. Visualisations work par-

ticularly well with tree-based machine learning algo-

rithms such as decision trees and random forests (An-

gelini et al., 2017), (Sopan et al., 2018), (Nadeem

et al., 2021). Other approaches include providing vi-

sual analytics regarding the data input, so that human

experts can perform manual investigation in further

(Angelini et al., 2017), or presenting additional visual

information selectively based on the human expert’s

trust levels (Anjomshoae et al., 2019).

2.3.2 XAI Enabled by Usability Testing

When it comes to effective, accurate, and correct de-

cision making by human experts, usability becomes

a priority. Complex visualisations could overwhelm

human experts rather than assisting them (Nadeem

et al., 2022). Therefore, the explanations provided by

an XAI model must be as clear, precise, and simplistic

as possible so human experts can effectively and efﬁ-

ciently utilise them (Antwarg et al., 2021), (Panigutti

et al., 2022). There is a also a need to eliminate the

misconception that explainability automatically pro-

vides interpretability (Nadeem et al., 2022). Inter-

pretability can improve decision making since it can

identify bias in the training data and can also ensure

that the variables involved meaningfully contribute to

the results of the AI model (Arrieta et al., 2020).

Additionally, achieving understandability in-

volves ensuring multiple usability factors that are

evaluated through effective user testing (Doshi-Velez

and Kim, 2017). Hence, it is essential for an XAI

model to be user-centric and multiple rounds of user

and usability testing to be conducted (Doshi-Velez

and Kim, 2017). Sepciﬁcally, before an XAI model

is deployed on top of the AI system in place, in-depth

usability testing must be conducted. In this way, en-

lightening feedback will be received from the users

and all the necessary changes will be made. Essen-

tially, an XAI model should be designed and evalu-

ated based on the feedback from the users involved.

2.3.3 XAI Enabled by Different Stakeholders

The results generated by an AI model are essen-

tial for effective decision-making procedures by var-

ious types of stakeholders exhibiting different back-

ground knowledge, expertise, experience, needs, and

motivates. Hence, each stakeholder will potentially

require different understandability, comprehensibil-

ity, interpretability, explainability, and transparency

approaches, tailored and adapted to their individual

needs (Blumreiter et al., 2019).

3 XAI IN FOOTBALL

XAI aims to support in accurate, effective and correct

decision making (Nadeem et al., 2022). In football in-

dustry, there are multiple types of users/stakeholders

involved in order for a team to be fully functional.

Examples include managers, assistant managers, ﬁt-

ness/conditioning coaches, chief sports analysts, nu-

tritionists, physiotherapists, and scouters. Every user

comes from a different scientiﬁc background, with

different levels of experience and expertise, respon-

sibilities and roles in the team. In practice, an XAI

model should help each of these users understand how

AI models were constructed, function, what the re-

sults reveal, and provide useful insights and conclu-

sions about the input data it received, tailored to their

needs, level of experience and knowledge (Nadeem

et al., 2022). We proceed by discussing how XAI

could be integrated in various football aspects, high-

lighting which of the characteristics and goals deﬁned

in the previous section are vital for its effective utili-

sation. Our guidelines and recommendations are also

summarised in Table 1.

3.1 XAI in Scouting

Recruiting/scouting is one of the most important pro-

cesses for football clubs. Most of the clubs have a

combination of human scouting experts physically at-

tending games and watching the potential players to

be signed and also using AI tools to assess a potential

player’s suitability for the team.

Several stages in scouting can enhanced by AI.

The ﬁrst stage consists of the AI model identifying

which positions in the team require new/additional

players for maximised performance and effective

player-rotating. The second stage is collecting the rel-

evant data for accurate comparison between the play-

ers the team is interested in recruiting. This includes

performance statistics for each player, such as passes

(short/long/dangerous) completed, successful assists,

The 12th Player: Explainable Artiﬁcial Intelligence (XAI) in Football: Conceptualisation, Applications, Challenges and Future Directions

215

Table 1: Summary of XAI utilisation to the Football Industry.

Stakeholders XAI Application in Football

Players

-Injury prediction and prevention

-Personalised rehabilitation programmes

-Personal training programme

Managers

Assistant Managers

- Decision making at player and team level

- Exploration of alternative tactics

- Player position changes for improved team performance

Coaches

Fitness/Training Experts

- Youth/Pro Players monitoring and development

- Performance optimisation

Sports Analysts

Chief Sports Analysts

-Tactical/Opposition analysis

-Gameplay patterns

Nutritionists

Physiotherapists

Healthcare Specialists

Medical Team

- Injury risk monitoring and Targeted interventions

-Rehabilitation monitoring and adjustments

-Wellbeing and fatigue level monitoring

-Physical and mental health monitoring

Scouters - Talent identiﬁcation

on and off target shots, successful dribbles, successful

aerial battles won, successful interceptions/tackles,

successful saves (goalkeepers), and so on. The AI-

informed scouting team can subsequently go and ac-

tually watch the players in-action to gather subjec-

tive evidence, beyond statistics (Ryan Beal and Ram-

churn, 2019). Examples include but not limited to

off-the-ball movement abilities, positioning, bravery

in pursuing challenges, team responsibility, stamina,

concentration, good communication and good re-

ﬂexes. Once all the necessary data is gathered and

based on the available budget, optimal decisions can

be taken. This is essentially an AI-optimised prob-

lem, meaning getting the most highly-rated players

with the least budget. There are also other factors to

consider, such as squad sizes and player wage caps.

As the literature suggest (Ryan Beal and Ramchurn,

2019), scouting is a process that can involve both hu-

man experts and AI models, each bringing their own

skills in the formula.

XAI could assist in the scouting process by ﬁrstly

identifying the best players for the required posi-

tions, why these players were selected and based on

which factors (understandability and explainability),

and how these factors contributed to their selection

(interpretability). Moreover, the XAI model, based on

the observations and feedback provided by the human

experts (scouters), can visually quantify which play-

ers are the most suitable and why (understandability

and explainability) and what requirements they sat-

isfy (interpretability). Finally, XAI could suggest on

alternative players, based on available budget.

3.2 XAI in Tactical Analysis

Unarguably, football is one of the most complex

sports and one of the most challenging to analyse due

to the large number of players involved and their var-

ied roles. AI can model the behaviour of players and

identify team gameplay patterns (Moustakidis et al.,

2023), in an attempt to provide sufﬁcient information

on which players are the most inﬂuential on the pitch,

the most reliable when it comes to goal scoring, the

most successful in stopping counter-attacks, with the

highest number of dangerous passes, and so on.

XAI could assist in explaining how different for-

mations of the squad could contribute to the play-

ers’/team’s improved performance (understandabil-

ity), thus explaining why certain players perform bet-

ter under certain circumstances (different position,

opponent, cooperation) (explainability and compre-

hensability) and what factors (features) have con-

tributed to it, as well as how these factors were han-

dled from the AI model (interpretability). Moreover,

based on what the manager and assistant managers

would like to try out, XAI could be used to suggest

alternative positions for certain players and different

formations, where their performance is maximised. In

addition, XAI could similarly used to analyse the op-

position team. In that way, the manager and the assis-

tant manager can gain valuable and insights on what

they are up against and make improvements to their

team and/or construct alternative strategies.

icSPORTS 2023 - 11th International Conference on Sport Sciences Research and Technology Support

216

3.3 XAI in Player Development and

Performance Optimisation

Football clubs around the world invest a lot on their

youth academies. Therefore, the appropriate develop-

ment of youth players is essential for becoming suc-

cessful candidates for joining the ﬁrst team squad.

AI systems can assist and optimise this process in

numerous stages. Firstly, AI can provide informa-

tion on the players’ performance both during practice

and game days. Input data can be performance re-

lated (e.g., successful shots/assists, passes completed,

tackles/interventions, etc.) or physical data captured

from GPS vests (e.g., total distance, high intensity

distance, sprint distance, top speeds, etc.) (Ryan Beal

and Ramchurn, 2019). Moreover, AI can be vi-

tal in personalising the training programme for each

footballer, according to their skills and capabilities

and the needs of the team. AI could also assist in

decisions relating to assigning a player to a higher

level team or a smaller club for gaining more expe-

rience and playing minutes. More speciﬁcally, an

AI model could map players (youth/professional) to

speciﬁc strength/conditioning programmes based on

their physical characteristics and objectives set by the

coaching team. XAI could assist in visually quanti-

fying the performance of players (understandability),

thus explaining why a player is performing in a spe-

ciﬁc way (explainability and comprehensability) and

what factors contributed, as well as how these factors

were handled from the AI model (interpretability).

In this way, the ”weak points” of a player can

be identiﬁed allowing the personnel involved to take

informed action. In case a footballer needs to be-

come physically stronger, the strength and condition-

ing team can re-construct their ﬁtness programme

along with the nutritionist who can improve their diet

plan. If the footballer needs to work on improving

their technique and/or speciﬁc actions (e.g. passing

accuracy), the manager/coaching personnel could use

re-construct their individual practice programme, so

they can improve. In addition, XAI could suggest

alternative positions for certain players (especially

youth players), based on their performance and phys-

ical characteristics (understandability, explainability

and comprehensability) and how certain factors con-

tribute to the alternative position (interpretability).

Managers and assistants could use this information to

improve their team tactics/strategies and rotation sys-

tems, coaches could differentiate their personal pro-

grammes, while the conditioning team could make the

necessary changes to their exercise programme based

on their role on the pitch.

3.4 XAI in Injury Prediction/Prevention

Injuries in football can be severe and many times sea-

son ending. Certain types of injuries may be avoided

or at least their magnitude and impact may be signif-

icantly minimised. AI could be used as an injury pre-

diction and prevention tool by monitoring the players’

overall movement to identify poor body posture, in-

correct form or overexertion, allowing potential risks

of injury to be identiﬁed. The AI model could make

insightful suggestions to readjust posture or to pre-

vent an injury, such as speciﬁc exercise programmes

to strengthen certain muscle groups. XAI could be

used to identify the types of injuries each player is at

risk of experiencing, visualising the factors contribut-

ing to this risk (understandability and explainability)

and how these factors can impact the players’ health

and wellbeing (interpretability) through visual quan-

tiﬁcation of these risks. The results obtained can help

the coaching and medical teams understand what ad-

justments need to be made and how, hence creating

personalised exercise programmes to ensure the play-

ers’ health, safety, and improved performance.

In addition, the game minutes could be analysed

by the AI model, to estimate which players are poten-

tially at risk of injuring themselves based on the num-

ber games played and their resting periods between

games. Once again, the XAI model could provide ac-

curate visualisations on the potential ”fatigue” levels

of the players and how these are constructed. Such

results can assist the managers into constructing an

optimised and safer player-rotation system.

3.5 XAI in Rehabilitation

Inevitably, most footballers will suffer from injuries

during their career. Some of these injuries may be

long-term and require one or more surgeries. Hence,

proper rehabilitation is essential for effective recov-

ery. Undeniably, every footballer is different and

therefore, every injury in manifested differently. AI

could assist in personalising the rehabilitation pro-

gramme for each player based on the type and sever-

ity of injury, their biomechanics and how each player

corresponds to the rehabilitation programme to max-

imise the probability of a successful recovery. In de-

tail, the AI model could take a closer look to each

player’s biomechanics so it can spot any weaknesses

or imbalances that may contribute to their slower re-

habilitation. In addition, AI could estimate approxi-

mate recovery timelines based on the current progress

made by the player as well as similar previous cases.

Finally, AI could track players’ movements in real-

time and provide effective feedback to maximise their

The 12th Player: Explainable Artiﬁcial Intelligence (XAI) in Football: Conceptualisation, Applications, Challenges and Future Directions

217

progress and minimise the risk of re-injury.

Through XAI, medical/healthcare special-

ists (sports injury rehabilitation, physiother-

apy/conditioning team) can understand the purpose of

a speciﬁc rehabilitation programme (understandabil-

ity, explainability) and what factors contributed to its

selection. Examples of important factors include the

type/severity of injury, medical intervention (if any),

whether it is a recurrent injury, player’s biological

characteristics (e.g., age, height, weight, BMI),

biomechanics (e.g. muscle imbalances) and perfor-

mance results (e.g., stamina levels). In addition, XAI

could explain why certain movements put the player

at risk of re-injury by visualising how they currently

move and how they should move correctly to avoid

putting additional physical stress to the injured area.

XAI should also assist in presenting similar past

cases (similar injuries/recovery) to illustrate common

points between them and the current case to support

its decision (Comprehensability). Interpretablity

could provide more information on how these factors

are used by AI and how inﬂuential and important

they were when choosing the speciﬁc rehabilitation.

Additionally, managers and assistant managers

can keep track of their players’ progress and get an

estimation on their possible return date. In addition,

nutritionists could be informed though the XAI model

on the players’ current exercise programme and its

physical requirements, so they could construct a cor-

responding diet plan based on their bodies needs. Fi-

nally, transparency should be considered so that every

player is treated fairly with regards to their condition,

receives clear instructions on what is required from

them throughout the different stages of the rehabilita-

tion programme and what improvements can be made

from their side to maximise results.

4 CHALLENGES AND FUTURE

DIRECTIONS

XAI is an interdisciplinary research ﬁeld in the

realm of AI (Brunotte et al., 2022),(Langer et al.,

2021a),(Langer et al., 2021b). Hence, several chal-

lenges emerge when attempting to provide sufﬁcient

explanations on its functioning. Firstly, providing

clear, precise, and sufﬁcient explanations is more

complex than it seems. It is essentially a sociologi-

cal issue (Miller, 2019), which ultimately refers to the

following question: how can we evaluate the qual-

ity of an explanation? (Longo et al., 2020). Ulti-

mately, there is no single correct answer as it can have

multifaceted implications for different stakeholders.

Therefore, in-depth cognitive experiments should be

conducted to decide on the presentation of explana-

tions (Longo et al., 2020). This is demonstrated in the

football case study presented, as multiple stakehold-

ers are involved with different background knowl-

edge, expertise, experience, motives, goals, and re-

sponsibilities. Therefore, what comprises an explain-

able set of information can vary for each stakeholder.

A possible approach towards simpliﬁed and straight-

forward explainability would be to use abstractions

(Gunning et al., 2019). In addition, an XAI model

should be able to explain to human experts about

not only its decisions but also its skills and capabil-

ities (Gunning et al., 2019). Even the most sophisti-

cated AI systems have their own blind spots and may

have cases which they might not have the perfect an-

swer. An XAI model should harmonically co-exist

with human experts to develop cross-disciplinary in-

sights and common baselines, helping both human ex-

perts and other AI agents learn, and improving their

own knowledge by using past knowledge and experi-

ence to accelerate discovery. This re-emphasises the

human aspects of AI.

Furthermore, one should consider the legal impli-

cations and the factor of accountability when it comes

to XAI. If an XAI model provides an incorrect sug-

gestion, what are the legal implications? Who is af-

fected? What are the consequences? What are the

socio-economical, health and safety implications? In

addition, we should consider whether XAI explana-

tions could negatively impact the safety and privacy

of individuals (et. al., 2023). Some important ques-

tions to ask are: How can XAI impact privacy and

safety? Which human factors should be considered?

Such questions are especially important when sensi-

tive information is involved. In the present case study,

this may relate to the players’ injuries records, per-

formance and medical data. Even an accidental leak-

age, could lead to great consequences to individuals

and teams. A possible countermeasure would be to

anonymise data and partially release anonymised sen-

sitive data as a baseline to collect more intelligence

about the individuals involved (Cai et al., 2016).

Moreover, XAI should be human-centred to facili-

tate usable and beneﬁcial interactions between human

experts and AI (Amershi et al., 2019). Beyond the rel-

evant HCI-related principles that should deﬁnitely be

considered when designing an XAI-enabled interface,

multiple iterations of user testing and usability studies

should be conducted. In addition, important questions

should be discussed before proceeding with designing

the XAI-enabled interface (Bush et al., 1945), such as

whether XAI should include additional explanations

on particular users who may lack relevant knowledge,

and how explanations can become more interactive to

icSPORTS 2023 - 11th International Conference on Sport Sciences Research and Technology Support

218

keep human users engaged while assisting them in un-

derstanding difﬁcult concepts more efﬁciently. In the

case study presented, it is clear that deploying a one-

ﬁt-for-all XAI system is not sufﬁcient and that the ap-

proach should be differentiated based on the individ-

ual’s knowledge, skills, experience and duties.

In addition, XAI should contribute towards more

responsible AI. It is no secret that an AI system au-

tomates multiple and complex tasks simultaneously.

Although its capabilities are remarkable with how fast

and accurate considerable amounts of data can be

handled, the fact remains that errors and miscalcula-

tions will be made. In such cases, it should be clear

and precise on how such mistakes are handled, who is

responsible to identify and respond to any mistakes,

who should be held accountable (et. al., 2023).

5 CONCLUDING REMARKS

The effective and correct conceptualisation and utili-

sation of XAI is still a work in progress. This holds

especially true for the football (soccer) sector, and

the sports industry in general, as to the best of our

knowledge no substantial research has been made for

it yet. Hence, in this paper we aimed to discuss how

XAI could be applied in various domains of a foot-

ball club to beneﬁt all the relevant stakeholders. In

particular, we have discussed how XAI is deﬁned and

what its characteristics are and the need for XAI to be

user-centred. We explored the domains of the foot-

ball industry (focusing on the footballers-related pro-

cedures) where XAI can be utilised and how can this

be beneﬁcial for the diverse target groups involved.

Finally, since XAI is a relatively recent area of re-

search, there is still plenty of improvement and work

to be done. Hence, we discuss what challenges XAI

faces, focusing on the case presented. In the future,

it is essential for XAI experts to address these chal-

lenges so XAI can be deployed alongside AI. Without

a doubt, XAI is going to be one of the most important

components an AI system should possess. As the ﬁeld

of AI expands its application areas and obtains a more

substantial social role, it is essential to remain in-sync

to the needs of human users.

REFERENCES

Adadi, A. and Berrada, M. (2018). Peeking inside the

black-box: a survey on explainable artiﬁcial intelli-

gence (xai). IEEE access, 6:52138–52160.

Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi,

B., Collisson, P., Suh, J., Iqbal, S., Bennett, P. N.,

Inkpen, K., et al. (2019). Guidelines for human-ai in-

teraction. In Proceedings of the 2019 chi conference

on human factors in computing systems, pages 1–13.

Angelini, M., Aniello, L., Lenti, S., Santucci, G., and Ucci,

D. (2017). The goods, the bads and the uglies: Sup-

porting decisions in malware detection through visual

analytics. In 2017 IEEE Symposium on Visualization

for Cyber Security (VizSec), pages 1–8. IEEE.

Anjomshoae, S., Najjar, A., Calvaresi, D., and Fr

amling,

K. (2019). Explainable agents and robots: Results

from a systematic literature review. In 18th Interna-

tional Conference on Autonomous Agents and Multia-

gent Systems (AAMAS 2019), Montreal, Canada, May

13–17, 2019, pages 1078–1088. International Founda-

tion for Autonomous Agents and Multiagent Systems.

Antwarg, L., Miller, R. M., Shapira, B., and Rokach, L.

(2021). Explaining anomalies detected by autoen-

coders using shapley additive explanations. Expert

systems with applications, 186:115736.

Arrieta, A. B., D

ıaz-Rodr

ıguez, N., Del Ser, J., Bennetot,

A., Tabik, S., Barbado, A., Garc

ıa, S., Gil-L

opez, S.,

Molina, D., Benjamins, R., et al. (2020). Explainable

artiﬁcial intelligence (xai): Concepts, taxonomies, op-

portunities and challenges toward responsible ai. In-

formation fusion, 58:82–115.

Bharati, S., Mondal, M. R. H., and Podder, P. (2023). A re-

view on explainable artiﬁcial intelligence for health-

care: Why, how, and when? IEEE Transactions on

Artiﬁcial Intelligence.

Bharati, S., Podder, P., and Mondal, M. (2020a). Artiﬁcial

neural network based breast cancer screening: a com-

prehensive review. arXiv preprint arXiv:2006.01767.

Bharati, S., Podder, P., and Mondal, M. R. H. (2020b). Hy-

brid deep learning for detecting lung diseases from

x-ray images. Informatics in Medicine Unlocked,

20:100391.

Biran, O. and Cotton, C. (2017). Explanation and justiﬁ-

cation in machine learning: A survey. In IJCAI-17

workshop on explainable AI (XAI), volume 8, pages

8–13.

Blumreiter, M., Greenyer, J., Garcia, F. J. C., Kl

os, V.,

Schwammberger, M., Sommer, C., Vogelsang, A.,

and Wortmann, A. (2019). Towards self-explainable

cyber-physical systems. In 2019 ACM/IEEE 22nd In-

ternational Conference on Model Driven Engineer-

ing Languages and Systems Companion (MODELS-

C), pages 543–548. IEEE.

Brunotte, W., Chazette, L., Kl

os, V., and Speith, T. (2022).

Quo vadis, explainability?–a research roadmap for ex-

plainability engineering. In International Working

Conference on Requirements Engineering: Founda-

tion for Software Quality, pages 26–32. Springer.

Bush, V. et al. (1945). As we may think. The atlantic

monthly, 176(1):101–108.

Cai, Z., He, Z., Guan, X., and Li, Y. (2016). Collective data-

sanitization for preventing sensitive information infer-

ence attacks in social networks. IEEE Transactions on

Dependable and Secure Computing, 15(4):577–590.

Chazette, L., Brunotte, W., and Speith, T. (2021). Exploring

explainability: a deﬁnition, a model, and a knowledge

The 12th Player: Explainable Artiﬁcial Intelligence (XAI) in Football: Conceptualisation, Applications, Challenges and Future Directions

219

catalogue. In 2021 IEEE 29th international require-

ments engineering conference (RE), pages 197–208.

IEEE.

Doshi-Velez, F. and Kim, B. (2017). Towards a rigorous sci-

ence of interpretable machine learning. arXiv preprint

arXiv:1702.08608.

Duell, J., Fan, X., Burnett, B., Aarts, G., and Zhou, S.-M.

(2021). A comparison of explanations given by ex-

plainable artiﬁcial intelligence methods on analysing

electronic health records. In 2021 IEEE EMBS Inter-

national Conference on Biomedical and Health Infor-

matics (BHI), pages 1–4. IEEE.

et. al., O. G. (2023). Six human-centered artiﬁcial intel-

ligence grand challenges. International journal of

human-computer interaction, 39(3):391–437.

Goodman, B. and Flaxman, S. (2017). European union reg-

ulations on algorithmic decision-making and a “right

to explanation”. AI magazine, 38(3):50–57.

Gunning, D., Steﬁk, M., Choi, J., Miller, T., Stumpf, S.,

and Yang, G.-Z. (2019). Xai—explainable artiﬁcial

intelligence. Science robotics, 4(37):eaay7120.

Gupta, A., Anpalagan, A., Guan, L., and Khwaja, A. S.

(2021). Deep learning for object detection and scene

perception in self-driving cars: Survey, challenges,

and open issues. Array, 10:100057.

ohl, M. A., Baum, K., Langer, M., Oster, D., Speith, T.,

and Bohlender, D. (2019). Explainability as a non-

functional requirement. In 2019 IEEE 27th Inter-

national Requirements Engineering Conference (RE),

pages 363–368. IEEE.

Langer, M., Baum, K., Hartmann, K., Hessel, S., Speith,

T., and Wahl, J. (2021a). Explainability audit-

ing for intelligent systems: a rationale for multi-

disciplinary perspectives. In 2021 IEEE 29th inter-

national requirements engineering conference work-

shops (REW), pages 164–168. IEEE.

Langer, M., Oster, D., Speith, T., Hermanns, H., K

astner,

L., Schmidt, E., Sesing, A., and Baum, K. (2021b).

What do we want from explainable artiﬁcial intel-

ligence (xai)?–a stakeholder perspective on xai and

a conceptual model guiding interdisciplinary xai re-

search. Artiﬁcial Intelligence, 296:103473.

Longo, L., Goebel, R., Lecue, F., Kieseberg, P., and

Holzinger, A. (2020). Explainable artiﬁcial intelli-

gence: Concepts, applications, research challenges

and visions. In International cross-domain conference

for machine learning and knowledge extraction, pages

1–16. Springer.

Miller, T. (2019). Explanation in artiﬁcial intelligence: In-

sights from the social sciences. Artiﬁcial intelligence,

267:1–38.

Mittelstadt, B., Russell, C., and Wachter, S. (2019). Ex-

plaining explanations in ai. In Proceedings of the con-

ference on fairness, accountability, and transparency,

pages 279–288.

Mondal, M. R. H., Bharati, S., and Podder, P. (2021). Co-

irv2: Optimized inceptionresnetv2 for covid-19 detec-

tion from chest ct images. PloS one, 16(10):e0259179.

Moustakidis, S., Plakias, S., Kokkotis, C., Tsatalas, T.,

and Tsaopoulos, D. (2023). Predicting football team

performance with explainable ai: Leveraging shap to

identify key team-level performance metrics. Future

Internet, 5(5):174.

Nadeem, A., Verwer, S., Moskal, S., and Yang, S. J.

(2021). Alert-driven attack graph generation using s-

pdfa. IEEE Transactions on Dependable and Secure

Computing, 19(2):731–746.

Nadeem, A., Vos, D., Cao, C., Pajola, L., Dieck, S., Baum-

gartner, R., and Verwer, S. (2022). Sok: Explainable

machine learning for computer security applications.

arXiv preprint arXiv:2208.10605.

Nazar, M., Alam, M. M., Yaﬁ, E., and Su’ud, M. M. (2021).

A systematic review of human–computer interaction

and explainable artiﬁcial intelligence in healthcare

with artiﬁcial intelligence techniques. IEEE Access,

9:153316–153348.

Padilla, L. M., Creem-Regehr, S. H., Hegarty, M., and Ste-

fanucci, J. K. (2018). Decision making with visualiza-

tions: a cognitive framework across disciplines. Cog-

nitive research: principles and implications, 3(1):1–

25.

aez, A. (2019). The pragmatic turn in explainable artiﬁcial

intelligence (xai). Minds and Machines, 29(3):441–

459.

Panigutti, C., Beretta, A., Giannotti, F., and Pedreschi, D.

(2022). Understanding the impact of explanations on

advice-taking: a user study for ai-based clinical deci-

sion support systems. In Proceedings of the 2022 CHI

Conference on Human Factors in Computing Systems,

pages 1–9.

Rathi, K., Somani, P., Koul, A. V., and Manu, K. (2020).

Applications of artiﬁcial intelligence in the game of

football: The global perspective. Researchers World,

11(2):18–29.

Russell, S. J. (2010). Artiﬁcial intelligence a modern ap-

proach. Pearson Education, Inc.

Ryan Beal, T. J. N. and Ramchurn, S. D. (2019). Artiﬁcial

intelligence for team sports: a survey. The Knowledge

Engineering Review, 34:1–40.

Sopan, A., Berninger, M., Mulakaluri, M., and Katakam,

R. (2018). Building a machine learning model for the

soc, by the input from the soc, and analyzing it for the

soc. In 2018 IEEE Symposium on Visualization for

Cyber Security (VizSec), pages 1–8. IEEE.

Van Lent, M., Fisher, W., and Mancuso, M. (2004). An ex-

plainable artiﬁcial intelligence system for small-unit

tactical behavior. In Proceedings of the national con-

ference on artiﬁcial intelligence, pages 900–907. Cite-

seer.

Vishwarupe, V., Joshi, P. M., Mathias, N., Maheshwari, S.,

Mhaisalkar, S., and Pawar, V. (2022). Explainable ai

and interpretable machine learning: A case study in

perspective. Procedia Computer Science, 204:869–

876.

Viton, F., Elbattah, M., Gu

erin, J.-L., and Dequen, G.

(2020). Heatmaps for visual explainability of cnn-

based predictions for multivariate time series with ap-

plication to healthcare. In 2020 IEEE International

Conference on Healthcare Informatics (ICHI), pages

1–8. IEEE.

icSPORTS 2023 - 11th International Conference on Sport Sciences Research and Technology Support

220