Analyzing Momentum Dynamics in Matches: A Data-Driven Approach for Performance Enhancement

Abstract

This research delves into the intricate dynamics of momentum in tennis matches by developing a sophisticated mathematical model. By employing a BP neural network for pre-training to address missing data points in continuous datasets, the study achieves significantly enhanced accuracy, as evidenced by an RMSE value of 21.05 on the test set, surpassing traditional imputation methods. In addition, by analysing the importance of the features, key features such as the running distance of the athletes were identified, which provided a scientific basis for improving the athletes' competition level. By applying these important features to the Adaboost machine learning model, this study achieved significant results in accurately identifying the key performance features of athletes, achieving a prediction accuracy of up to 0.957 and accurately predicting the trend of the game. The model's successful application across diverse race scenarios in various regions, genders, and years highlights its robust generalizability and potential for widespread practical use. The significance of this research lies in its methodological innovation and practical applications. By employing advanced machine learning techniques to fill in missing data and predict performance trends, it offers a novel approach to understanding and enhancing athletic performance.