Research on momentum quantification and influencing factors based on machine learning

From the score data of the Wimbledon tennis matches, we observed unexpected score fluctuations and even set-level variations for the leading side. This article aims to evaluate which player performs better at any given moment during the match and quantify the extent of their performance advantage, which this paper refers to as momentum. Firstly, this paper defines momentum as the player's winning probability, convert match data into a Markov chain, and establish a time-predictive model based on the score timeline using Hidden Markov Models. Upon importing match data into our model, this paper successfully computes the magnitude of momentum for each player. Subsequently, this paper utilizes exploratory factor analysis, linear regression analysis, and Pearson correlation coefficient analysis to calculate the degree of association between momentum impact factors and player performance, observing that variables such as serve_no, point_victor, p1_winner, winner_shot_type, p1_net_pt_won, p1_distance_run, rally_count, and speed_mph exhibits strong correlations with momentum. For example, higher serve speeds are likely to increase a player's chance of winning, but break points faced, unforced errors, and distance run by the player may reduce the overall probability of a player winning. When applying the model output in actual matches, coaches and players can formulate strategies for serving selection and tactical arrangements based on the information provided by the model to maximize the utilization of momentum changes and develop targeted strategies against opponents' weaknesses and habits.