A Study on Momentum Prediction for Tennis Matches Based on Sliding Window Algorithm and LSTM Networks

In this paper, the influence of momentum in tennis matches is studied, and the sliding window algorithm and LSTM model are used to evaluate and predict the performance of players during the match. First, a sliding window-based weighting mechanism is used to define momentum, taking the score situation within the window as the key measure and giving more weight through logical functions. Secondly, based on the Kolmogorov-Smirnov test, the lack of correlation between the random simulation and the actual results is proved. At the same time, the existence of momentum is proved by comparing the defined momentum trend with the actual course of the match. Finally, by using a democratic voting system based on XGBoost, Mutual Information and Random Forest, the nine variables most relevant to the contest process were selected. These variables are used to train our LSTM neural network model, and tests show that the model has strong generalization and robustness.