Tennis Match Win Prediction Based on RF and RNN

This study first preprocessed the data and constructed new indicators based on existing data to better describe the changes that occurred during the game. Considering the complexity of tennis matches, and constructed new data sets according to different time scales and tried to construct models based on the characteristics of different time scales. For the set scale, this study constructed a win-loss prediction model based on random forest to overall measure the player's performance. For the point scale, this study established a winning rate classification prediction model based on recurrent neural networks. On this basis, considering the higher impact of having the right to serve on the winning rate, the overall average winning rate of serving first was 0.673 based on the Bayesian formula. We construct a four-class classification model of recurrent neural network winning rate based on Bayesian probability. The model is evaluated through indicators such as confusion matrix, ROC curve and AUC value. Finally, taking game 1701 as an example, the game trends and player performance were analyzed, and the corresponding probabilities under four different categories and the final player's winning situation were obtained to reflect the excellence of the model.