A Data-Driven Approach to Tennis Player Performance Prediction and Evaluation

Abstract

This study presents a comprehensive framework for evaluating and predicting tennis player performance using historical match data. First, to reduce the data dimension, principal component analysis is used to obtain seven principal components that capture more than 85% of the variance of the data. These principal components serve as inputs to a machine-learning model that is used to predict the outcome of tennis scores. Various machine learning algorithms are compared using ROC curves and evaluation metrics. The gradient gradient-enhanced decision tree (GBDT) model performs best. To evaluate athletes' performance comprehensively, an evaluation system based on the derived principal component was constructed by using the entropy weight method and the TOPSIS model. This comprehensive approach achieved 89.04% accuracy in predicting the outcome of tennis matches, demonstrating its effectiveness in capturing match dynamics. In addition, the study introduces a visualization method that combines macro-level score analysis with micro-level performance assessment. This allows a detailed understanding of how players perform at different stages of the game, enabling coaches and players to make informed decisions and optimize strategies.