LGBM-Based Real-Time Analysis of Tennis Player Performance

Abstract

This study aims to develop a real-time performance evaluation model for tennis players, enabling the quantification of match dynamics and the identification of key factors influencing scoring outcomes. Utilizing data from the Wimbledon Open dataset, the research applies preprocessing techniques such as outlier removal to construct a two-level indicator system encompassing psychological, physiological, and service-related variables. A binary logistic regression analysis is conducted to test indicator significance, followed by a comparative evaluation of five machine learning algorithms: LGBM, XGB, SVC, MLP, and LR. Results demonstrate that the LGBM algorithm outperforms others, achieving an accuracy of 0.69 and an AUC of 0.77. The model effectively captures real-time player performance and provides actionable data support for tennis training and tactical decision-making. Future improvements may focus on refining the indicator system to enhance the model’s generalizability and predictive robustness.