Machine Learning-Based Tennis Match Performance Prediction Model

The paper mainly builds a machine learning model for tennis match score prediction and conducts significance analysis on the effects of concomitant conditions using PSM. Firstly, the paper constructed a set of new metrics system, including whether the athlete is on the serve side, his/her personal skill, the level of fatigue, and his/her mentality during the match, tested the significant effect of these metrics on the model prediction by binary logistic regression, and used various machine learning model such as XGBOOST, SVC, LGBM to build the prediction model. Then, the paper trained an LGBM model based on the tennis match data set and improved the metrics system to realize the dynamic evaluation of players’ real-time performance, i.e., “momentum.” Then, in order to prove that players’ fluctuations and successes in the game are not random and that the role of "momentum" in the game really exists, the paper analyzes the concomitant conditions of scoring fluctuations using a counterfactual analysis framework and a propensity score matching algorithm and argues that the essential conditions have a significant effect on players’ performances by comparing the T-stat values of different core explanatory variables, which ultimately proves and further explains the importance of the role of "momentum" in the game’s outcome.