Prediction of substitution effectiveness in football: Brazilian Tournament First Division case

Abstract

The substitutions are essential for coaches to improve team performance. Besides, they can be crucial for the match result. The relevance and limitation of substitutions in official matches have conducted several studies to propose an optimal way to substitute a player. That is, the best moment during the match or the best strategy. Machine learning approaches are widely applied to build prediction models for improving the effectiveness of substitutions. Accordingly, this work proposes to build prediction models to help the away team coach deciding the best strategy and moment to replace a player during the match. We use a dataset extracted from four years of the Brazilian Tournament First Division Championship (2015-2018), and based on this dataset, we extract features from the matches and propose new ones (e.g., the team strength) to build the models. Four classifiers are applied: Support Vector Machine, Decision Tree, Random Forest, and K-Nearest Neighbor. The results show that the models can achieve up to 90% of F1-Score, and then they can be promising for analysis of the best moment to replace a player.