The elusive features of success in soccer passes: A machine learning perspective

Abstract

Machine learning has in recent years been increasingly used in the soccer realm. This paper focuses on investigating the factors influencing pass success, a chief element in team performance. Decision tree techniques are used aiming to identify which features are the most important in pass success. This process is applied to a data set of 13 matches of the men’s French “Ligue 1”. Two experiments are conducted using different feature sets: one containing the positional data and Voronoi area off all players, the second considering only the ball carrier and closest teammates and opponents. The results obtained with the first feature set indicate that the relative importance of features is match dependent and somehow related to teams’ formation and players’ tactical mission. The second feature set, being more directly related to the passing process, provided a more consistent ranking of features. Features related to the interaction with the opponent standout. Low precision and recall val ues show that the features and factors leading to pass success are in fact elusive.