TABLE 1.

Summary of studies using machine learning approaches exclusively with soccer players as their studied population.

Study	Participants Period of study	Main outcome studied	Main finding	Topic
Ayala et al. [5]	96 male professional players, 1 season	Hamstring strain injuries predicted	The prediction model showed moderate to high accuracy	Injury risk/Injury occurrence prediction
Oliver et al. [18]	355 elite youth players aged 10–18 years old, 1 season	Injuries predicted	The best performing decision tree model provided a specificity of 74.2% and sensitivity of 55.6% with an AUC of 0.663
Rossi et al. [17]	26 professional male players, 1 season	Non-contact injuries	Machine learning technique can detect around 80% of the injuries with about 50% precision, far better than the baselines and state-of-the-art injury risk estimation techniques
Rommers et al. [7]	734 players, under-10 to under-15 age categories, 1 season	Non-contact injuries	Machine learning algorithm was able to identify the injured players in the hold-out test sample with 85% precision, 85% recall (sensitivity) and 85% accuracy
Bongiovanni et al. [35]	16 male under-15 team players, 1 season	Sprint CoD, CMJ & aerobic fitness performance prediction	Anthropometric features were predictors of sprint performance and aerobic fitness, not CoD and CMJ	Physical performance prediction
Campbell et al. [29]	As the authors state "The data encompassed multiple seasons (2013–2018)and was pooled across pre-season and in-season training sessions” without including information on the data size.	Internal (sRPE) and external load (total distance covered)	Very low predictive ability	Players’ monitoring
Geurkink et al. [28]	46 elite male players, 61 training sessions, 913 observations	Predicted sRPE	sRPE was predicted accurately
Jaspers et al. [27]	38 professional male players, 2 seasons	External and internal training load	More accurate predictions of training Rate of Perceived Exertion from external workload data in combination with pre-session wellness
Op De Beéck et al. [26]	One soccer team (no information for duration of the study)	Future wellness items (i.e., fatigue, sleep quality, general muscle soreness, stress levels, and mood)	Wellness was predicted based on internal and external workload data. Their effect sizes indicate that the external load and internal load, separately and in combination, do not have sufficient predictive ability
Perri et al. [36]	28 sub-elite players, 1 season	Wellness index as predicted by internal training load	Machine learning technique predicted the wellness index based on previous training day internal load
Dick et al. [37]	Tracking data consisting of a sequence of coordinates of all players and the ball for a set of soccer games	Successful attacks	Proposed an approach to learn valuations of multiplayer positioning using positional data	Performance analysis
Goes et al. [23]	Position tracking data of 118 Dutch Eredivisie matches, containing 12424 attacks	Successful attacks	Identified dynamic formations based on position tracking data, and identified dynamic subgroups for every timeframe in a match
Link and Hoernig [24]	Data from 60 matches in the German Bundesliga, 1 season	Models for detecting individual and team ball possession based on position data	Match event were detected automatically
Montoliu et al. [25]	Football videos including two regular league matches played by up to four professional teams	Team activity recognition and analysis	The proposed method performed the team activity recognition task with high accuracy
Wang [8]	Teams playing in UEFA EURO2012	Technical and tactical analysis of teams	Key performance indicators were identified
Zago et al. [38]	13 elite female players performed a shuttle run test, wearing 6-axes inertial sensor at the pelvis level	Prediction of turn direction, speed (before/after turn) and the related positive/negative mechanical work	Good predictive ability of the machine learning algorithms	Movement analysis
Barron et al. [39]	966 outfield male players, 1 season	Identify key performance indicators that influence player’s career status	Specific technical characteristics correctly predicted 78.8% of the players’ league status with a test error of 8.3%	Player’s career trajectory
Matesanz et al. [40]	Soccer players’ transfer network among 21 European first leagues between the seasons 1996/1997 and 2015/2016	Table rank, UEFA points	Clubs with the highest transfer spending achieve better performance	Club performance
Sahin and Erol [41]	Data of 236 soccer games, 1 season	Predict the attendance demand in European soccer games	A model was proposed to predict attendance with higher accuracy	Match attendance

Note: AUC: Area Under the Curve; CoD: change of direction; CMJ: countermovement jump; sRPE: session rating of perceived exertion