Quantification and analysis of second balls in soccer

Abstract

In soccer, second balls are crucial to control possession and create attacking chances, but have remained largely unexplored. In this thesis, a mathematical framework is created to identify, classify, and extract second balls from data. Building on this foundation, the novel Expected Second Ball Value (xSBV) model uses machine learning and Markov chains to estimate both the probability of winning a second ball and the likelihood that the following possession leads to a goal. Predictive models achieved a top-3 accuracy of 60% for second ball location and an ROC-AUC score of 0.79 for predicting the winning team. The key results highlighted specific areas to target for higher success rates and produced a ranking of players based on their second-ball winning ability. This thesis extends existing literature for second ball analysis, offering valuable applications for player evaluation and tactical decision-making.