Momentum Analysis of Tennis Matches Based on Logistic Regression

The paper first selected ten evaluation indicators such as first-serve scoring rate, double fault rate, and ace rate through a literature review. Then, using the entropy weight method, weights were assigned to each evaluation indicator. The top seven indicators with larger weights were selected as independent variables, with the pointing situation as the dependent variable, to establish a binary logistic regression model. The model achieved an accuracy rate of 94.3%. The paper multiplied the regression-derived weight matrix with the seven indicators, normalizing the resulting values to represent the probability of a player pointing probability. The slope of the pointing probability obtained in this way was defined as momentum. Based on the real-time changes in pointing probability during matches, the article calculated the average momentum for each match. By conducting a chi-square test between the direction (positive or negative) of the average momentum and the corresponding match results, with a significance level less than 0.001, the paper demonstrated a strong correlation between the two, suggesting that match outcomes could be predicted by calculating the average momentum.