Research on the Momentum of Tennis Players Based on the XGBoost Model and GA-RF Model

Abstract

In this paper, to better represent an athlete's performance using momentum, this paper proposed an indicator system to assess a player's performance at different stages of a match. The article tested this system using both LightGBM and XGBoost classification models. By comparing the accuracy, recall, and precision of the training sets for both models, this paper found that LightGBM achieved 91.8% for all these metrics, while XGBoost achieved an impressive 98.4%. Therefore, this paper used the feature importance derived from the XGBoost model to determine the weight of these indicators. After obtaining the weighted indicators for the players, the essay used a comprehensive calculation formula to assess the momentum changes of a player during the match. The essay also applied a genetic algorithm-optimized random forest regression model (GA-RF) to predict real-time momentum changes in a match. Additionally, this paper utilized the SHAP model to interpret the predictions, providing specific recommendations for players to prepare for the match.