Unraveling the Secrets of Momentum Through GA-XGBoost

First, we build a model that applies to one or more games to identify key points and visualize the flow of the match. Then, it is necessary to analyze whether momentum will play a directional role in the direction and outcome of the race. To solve the problem of predicting the shift of the dominant side's players during a match, we define the psychology-determined momentum by multiple linear regression and the motion-determined momentum by linear regression. The entropy weighting method is used to calculate the weights of 60.797% and 39.203% for the psychological momentum and the movement momentum, respectively, and multiple linear regressions are used to finally define the momentum of each player and the inter-player momentum MD. Next, Kendall's W Test is performed using MATLAB software. The values of Kendall's Coefficient of Agreement for the two players are 0.871 and 0.884. It can be concluded that swing and success are not random, but have a very strong relationship. We treat this problem as a binary classification problem and first build a model using the GBDT algorithm to obtain the importance weights of each determinant. We introduce the regularization and loss functions to control the complexity of the model and improve the model to XGBoost. To improve the accuracy of the training and cross-validation sets of the model, we use a genetic algorithm (GA). The model test results show that the accuracy of the test set and cross-validation set are greatly improved. Our GA-XGBoost model is the best solution for this problem.