Quantifying and Predicting Momentum in Tennis Match via Machine Learning Approach

Abstract

This study aims to identify and analyze momentum shifts in tennis, developing a data-driven model to quantify and predict these shifts and assess their influence on match outcomes. Using data from 6 tournaments, including 564 matches and over 135,000 points, this study constructed a momentum calculation model integrating 14 weighted match factors such as point progression, server advantage, and player ranking differences. The model incorporates adjustments for set discontinuities and initial momentum based on player rankings to enhance predictive accuracy. Following data processing and validation, a Kappa consistency test was performed on the 2023 Wimbledon Championship data, yielding a high alignment with actual outcomes (Kappa = 0.96). Using a Gradient Boosting Decision Tree (GBDT) regression model, the study achieved a high accuracy in predicting momentum shifts, identifying key variables such as serve advantage and score gaps as primary indicators of performance dynamics. This model further revealed that players' momentum tends to stabilize at critical points, such as 40:30, while fluctuating more at disadvantageous scores. These findings highlight the model's utility for pre-match analysis, enabling detailed insights into opponents' tactical patterns and psychological responses under varying score conditions. Overall, this momentum model provides valuable applications for enhancing player preparation and in-game strategic adjustments, offering coaches and players a quantifiable tool to interpret and influence match outcomes.