A Study of Player Performance Prediction Based on Data-Driven LOGISTIC Test and Multi-Machine Algorithmic Modeling

In this paper, the index system based on players' individual technical ability, first-hand advantage, game mentality, fatigue level and offensive aggressiveness was constructed and the players' scores were evaluated and visualized. The significance test of the indicator system and players' scores was conducted through LOGISTIC to analyze the performance-related indicators. Based on the previously derived results, the Spearman correlation between the probability of a player's score and the actual score was examined and a linear regression model was constructed to explore the effect of momentum on a player's score. Then, the indicators were clustered to analyze the relationship between indicators and players' performance, the features were aggregated, the correlation between indicators was tested by LOGISTIC, and the MLP model was selected for modeling by 5-fold cross-validation. Finally, according to the importance of each indicator, reasonable suggestions were provided for players. Four games were randomly selected, and the “momentum” of the games was comprehensively measured by machine learning algorithms and visualized by ROC curves and AUC values. The technical strength of the competitors was factored into the future model. By analyzing the generalization performance of the model constructed by testing the 2015 Sudirman Cup match between Lin Dan and Tohma Ueda, it can be obtained that the model constructed above has good generalization performance for badminton, which may be due to the fact that there are many similarities between these two sports; the decrease in the model's performance may be due to the fact that tennis is more of a test of endurance while badminton is more of a test of explosive power. The reason for the decrease in model performance may be that tennis is more endurance-testing while badminton is more explosive-testing.