Mapping football tactical behavior and collective dynamics with artificial intelligence: A systematic review

José Eduardo Teixeira¹Eduardo Maio²Pedro Afonso²Samuel Encarnação³Guilherme Figueiredo Machado⁴Ryland Morgans⁵Tiago M. Barbosa³António Miguel Monteiro³Pedro Forte⁶Ricardo Ferraz⁷Luís Branquinho^2*

• ¹Instituto Politécnico da Guarda, Guarda, Guarda, Portugal
• ²Polytechnic Institute of Portalegre, Portalegre, Portalegre, Portugal
• ³Polytechnic Institute of Bragança (IPB), Bragança, Braganca, Portugal
• ⁴Universidade Federal de Viçosa, Viçosa, Minas Gerais, Brazil
• ⁵Cardiff Metropolitan University, Cardiff, United Kingdom
• ⁶Higher Institute of Educational Sciences of the Douro, Penafiel, Porto, Portugal
• ⁷University of Beira Interior, Covilhã, Castelo Branco, Portugal

Football, as a dynamic and complex sport, demands an understanding of tactical behaviours to excel in training and competition. Artificial intelligence (AI) has revolutionized the tactical performance analysis in football, offering unprecedented data analytics insights for players, coaches, and analysts. This systematic review aims to examine and map out the current state of research on AIbased tactical behaviour, collective dynamics and movement patterns in football. A total of 2,548 articles were identified following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and the Population-Intervention-Comparators-Outcomes (PICOS) framework. By synthesizing findings from 32 studies, this review elucidates the available AI-based techniques to analyze tactical behavior and identify collective dynamic based on Artificial Neural Networks (ANN), Deep Learning (DL), Machine learning (ML), and Time-Series (TS) techniques. Concretely, the tactical behavior was expressed by spatiotemporal tracking data using Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Variational Recurrent Neural Networks (VRNN), and Variational Autoencoders (VAE), Delaunay method, Player Rank, Hierarchical Clustering, Logistic Regression (LR), XGBoost, Random Forest (RF) Classifier, Repeated Incremental Pruning Produce Error Reduction (RIPPER), Principal Component Analysis (PCA), and T-Distributed Stochastic Neighbor Embedding (t-SNE). Furthermore, collective dynamics and patterns were mapped by graph metrics such as Betweenness centrality, Eccentricity, Efficiency, Vulnerability, Clustering Coefficient, and Page Rank, Expected Possession Value (EPV), Pitch Control Map classifier, Computer Vision techniques, Expected Goals (xG), 3D ball trajectories, Dangerousity Assessment (DA), Pass Probability Model (PPM ), and Total Passes Attempted (TPA). The technical-tactical key indicators performance (KPI) was expressed by team possession, team formation, team strategy, teamspace control efficiency, determining team formations, coordination patterns, analyzing player interactions, ball trajectories, and pass effectiveness. In conclusion, the AI-based models can effectively reshape the landscape of spatiotemporal tracking data into training and practice routines with real-time decision-making support, performance prediction, match management, tactical-strategic thinking, and training task design. Nevertheless, there are still challenges for the real practical application of AI-based techniques, as well as ethical regulation and the formation of professional profiles that combine sports science, data analytics, computer science, and coaching expertise.