Time Neural Network Tennis Match Prediction Model Based on Momentum Construction and Feature Extraction

People have widely discussed the prediction of the results of various sports competitions. However, the existing models are based on the historical data or the objective physical data of the game and do not consider the influence of the players' momentum on the result of the game. Therefore, this study used dataset from the 2023 Wimbledon Gentlemen’s singles matches to investigate the impact of momentum on the outcome of tennis matches. By developing a model that incorporates momentum values, the research aims to identify key factors influencing match results and improve predictive accuracy. Data preprocessing and feature selection were performed on a detailed point-by-point dataset. Specifically, Principal Component Analysis (PCA) was used to reduce dimensionality, followed by Random Forest to identify the most important features. By comparing with many machine learning models, our proposed model (LSTM-Momentum) shows excellent performance on several evaluation indicators, such as the accuracy (0.9770) and AUC (0.9995). The results show that integrating momentum values significantly enhances the model's ability to capture fundamental patterns and features, providing valuable insights for athletes, coaches, and sports researchers This study contributes to the understanding of momentum in tennis and offers a scientific basis for optimizing player training and strategy development.