Research on Predicting Momentum Changes in Tennis Matches Based on Markov and Random Forest Algorithms

Abstract

Nowadays, momentum is widely mentioned in sports competitions and is regarded as an important psychological and emotional state, but its essence and impact on the results of the game are difficult to accurately quantify and analyze. Therefore, studying the momentum effect in sports competitions and predicting momentum changes is particularly important. We first evaluate the momentum effect in the tennis match by calculating the state transfer matrix of the Markov chain about the tennis score. Using the data of 31 matches about 2023 Wimbledon men’s singles tennis match, we calculated that the server indeed has a higher probability of winning the serve game. In order to simplify the model, assuming that the transfer probability of the momentum state is equal to the transfer probability of the score state, we have established a momentum model that can capture the progress of tennis matches.Then, we have carried out a series of preprocessing of data, such as One hot encoding. Use a series of indicators of two players as independent variables to build a random forest model. The model has RMSE of 0.15, MAPE is 0.18, and  is 0.2. Then use the SHAP algorithm to find out the factors most relevant to the momentum of the game.