2025 22nd International Learning and Technology Conference (L&T) - 15-16 January 2025
Souheila Boudouda, Aya Merouani, Naila Marir
https://ieeexplore.ieee.org/document/10941377/authors#authors
The article presents a comprehensive AI-driven framework for analysing and predicting football player performance, integrating machine learning (ML), deep learning (DL), and recommendation systems to automate player evaluation and match tracking. It combines pre-match profiling, real-time match tracking, and post-match performance prediction into one system.
The study builds upon the evolution of AI in sports, tracing roots from Moneyball and data-driven talent evaluation to the present-day use of computer vision (YOLOv8), ML regressors (SVR, XGBoost, Decision Tree), and recommendation algorithms (cosine similarity). Its core objective is to create an intelligent system that predicts player performance and supports tactical decisions before, during, and after matches.
The research on the use of ML in Football, can be tracked to the early years of 2000’s, nevertheless, it is not until the last decade that tracking players’ performance took the initial steps into using ML, in order to track players’ performance during different cycles of a season. The main objective of using ML in tracking players performance is to have the ability to analyze and interpret large volumes of complex and mixed Data where ML algorithms can automatically identify correlations and physical attributes as an instance, and performance growth.
This study highlights the transformative potential of Artificial Intelligence when creating a data-driven framework for football performance analysis across pre-match, mid-match, and post-match contexts. It highlights how AI can unify traditionally isolated analytical processes such as player scouting, in-game tracking and post-match evaluation into a single, continuous feedback system that enhances tactical and performance-based decision-making.
The research demonstrates that the integration of supervised and unsupervised learning, particularly through clustering-regression methods, allows for more accurate predictions of player readiness and development. Similarly, the incorporation of AI-driven recommendation systems enables automated player comparisons, supporting evidence-based selection, substitution, and tactical planning. Another key learning concerns the application of deep learning, notably through YOLOv8, which facilitates real-time detection of players, referees, and match events. This development enhances the accuracy and efficiency of live performance monitoring and provides a technological foundation for in-game tactical analysis.
Relevance: Excellent (★★★★★)
This paper provides strong pre-match analytical insights by proposing a Player Profiling Module that uses machine learning to assess physical and technical attributes before a game.
By using a large dataset (19,000+ players, 89 attributes from Sofifa and Kaggle), the model performs clustering-regression to group players by ability and predict future performance trends while algorithms like SVR and XGBoost are used to estimate players’ physical readiness and progression, assisting coaches in lineup selection, tactical matching, and opponent preparation.
This framework makes pre-match planning more data-driven, replacing intuition with automated player similarity, versatility scoring, and position suitability recommendations.
The model is highly effective for pre-match use because it provides data-backed insight into player selection, squad composition, and opponent preparation.
This article highlights the real-time tracking using deep learning and how it works for mid-match analytics. Clubs can use YOLOv8 to detect and track players, referees and the ball and to identify key match events (corner kicks, goals). By combining tracking and event data, it enables the creation of live dashboards that visualise positional data, movements, and interactions during play and create models that can automate live tactical feedback, monitor player workload, and detect on-field trends without manual intervention.