Analysis of Momentum in Tennis Matches Based on Machine Learning Models

Momentum, as one of the critical factors affecting the outcomes of sports competitions, has long posed challenges in its quantification and mechanism of action. This study focuses on Wimbledon tennis matches, constructing and validating a multitask analytical framework to systematically explore momentum and its impact on match outcomes. Firstly, the study filters and normalizes predictive indicators affecting point wins and losses from Wimbledon match data. By combining binary logistic regression with various machine learning model optimizations, it designs an optimal model to capture scoring dynamics, enabling dynamic visual analysis of match processes. Secondly, momentum decomposition analysis is conducted based on data training results, and the potential influence of momentum on match outcomes is verified using the Kruskal-Wallis test (P = 0.27). The study further validates the model using data from the 2023 Wimbledon Men's Singles Final, achieving a predictive accuracy of 72% through aggregate processing and the Hosmer-Lemeshow test. A LightGBM model is employed to visualize momentum transitions, while analysis of variance (ANOVA) identifies key factors affecting momentum changes. Finally, the model's performance and potential error sources are evaluated using PRC curves and Pearson analysis. Comparative analyses across different sports validate the model's generalizability and cross-discipline applicability. The findings provide novel insights into the quantification of momentum and its application in match prediction, demonstrating the model's extensive potential for dynamic performance analysis in sports.