Uso de cluster para formação de equipes balanceadas na prática do futebol

Abstract

The use of data to support decision-making in sports has established itself as an important research and technological application field. In football, advances in the collection of individual performance metrics have enabled the use of machine learning methods for athlete analysis and sports management support. In informal and amateur contexts, however, team formation still occurs subjectively, often resulting in unbalanced matches. Against this backdrop, this work investigates the use of clustering techniques — with emphasis on hierarchical clustering with the Ward method — applied to player per formance data from the five main European leagues (2024/2025 season) as a tool for balanced team formation. The methodology followed a classical data science pipeline: data collection and consolidation; preprocessing; dimensionality reduction via Principal Component Analysis (PCA); clustering; and application of groupings for team composition. Two algorithms were compared — K-Means and Ward — evaluated through the Silhouette coefficient and the Davies-Bouldin index, as well as a football-based interpretive criterion and adherence to median formations observed in real teams. The Ward hierarchical solution with seven clusters demonstrated the best balance between functional refinement, tactical interpretability, and practical applicability, identifying profiles such as goalkeepers, ball-playing defenders, organizing midfielders, creative attackers, and finishers. The proposed system distributes players among teams considering both the functional diversity of clusters and the relative technical balance of athletes, outper forming simple average-based balancing strategies. The results indicate that clustering constitutes a promising and methodologically grounded approach for balanced team formation, with applicability in recreational, amateur, and semi-professional contexts.