Machine Learning-Based Analysis of Tennis Match Processes

Prediction is an important topic in machine learning. Given the random nature of sports events, they are meaningful experimental subjects for conducting research. In this paper, we focused on answering three key questions: (1) How to determine which player is performing better at a specific time? (2) What is the impact of so-called “momentum” on the outcome of a match? And (3) How to determine which player has the advantage at a certain point in the match. Firstly, we construct the wavelet neural network, in which Spearman's correlation coefficient is 0.7069, to decompose the performance of a player in real-time into relevant quantifiable features, evaluating which player is performing better at the moment. Then, we identified the main features affecting momentum and proposed the Additive-Multiplicative Model. The results confirm the momentum role in determining the outcome of a player's match. Finally, we established a random forest model with 30 decision trees. The accuracy and recall rates are 83.3% and 89.0%, respectively. Our study shows high accuracy and robustness in the outcome of sports events prediction and demonstrates promising potential for further applications.