Research on Data-Driven Deep Learning Models for Tennis Matches and Their Performance Optimization

This paper proposes an innovative deep learning model based on the combination of GCN and LSTM for predicting the outcome of tennis matches. This paper first constructs a dynamic relationship graph in the game, converts factors such as player interactions, scoring, and field status in the game data into nodes and edges of the graph, and captures the relationships and interactions between players through the GCN model. Then, the LSTM network is used to model the time series characteristics of the game to help analyze the dynamically changing game status during the game. Compared with traditional machine learning models, the model proposed in this paper can more effectively handle complex spatiotemporal dependencies, thereby improving prediction accuracy. A set of real tennis matches from different seasons was used to evaluate the model by means of precision, accuracy, recall, and F1. Experiments show that the prediction precision of this model is up to 90%, which is about 12% better than the conventional SVM and decision trees. Moreover, the optimized model has obvious superiority in accuracy and recall, which proves the validity of FNN and LSTM in the prediction of sports events.