Revealing the Close Connection Between Tennis Match Momentum and Victory Through Machine Learning Methods

Abstract

In competitive sports, especially in tennis matches, momentum is considered a key factor influencing the outcome. However, there is widespread controversy surrounding the definition of momentum, its quantification methods, and its impact on match results. We compared different models (LightGBM, SVM, XGBoost, MLP, logistic regression) using five-fold cross-validation to predict changes in momentum in tennis matches based on a given dataset. The results showed that LightGBM performed best with an accuracy of 77.6%. We selected LightGBM to predict the momentum score for each point of player 1 in the final match. It was found that the average momentum scores in the second, third, and fifth sets were greater than 0.5, which aligned with the actual outcomes. Furthermore, through Pearson correlation analysis and OLS validation of the relationship between momentum changes and match results, we obtained a p-value below 0.05, a Pearson correlation coefficient of 0.59, and an F-statistic of 376.6 in the OLS model, indicating a strong positive correlation between the two.