Exploring the dynamic performance of tennis matches: a comparison of momentum models based on neural networks and entropy weighting methods

Abstract

The aim of this paper is to develop two models to evaluate the performance of players in tennis matches, which these studies refer to as "momentum models". This paper described the process of building a momentum model, which included the building and solving of two models based on Artificial Neural Network (ANN) and Information Entropy Weighting Method (EWM). In the ANN-based model, the authors first constructed a neural network structure that could perfectly handle a given dataset and prove its effectiveness through comparative experiments. Then, the authors interpreted and validated the output results for the model to reflect its ability to describe the momentum of the race. In the EWM-based model, the authors determined the weights of the features through the information entropy weighting method, which were used to calculate the momentum of the players in the match, and demonstrated the changes of the momentum through linear and scatter plots. Subsequently, the authors compared and analyzed the momentum with the score and comprehensively evaluated the player's performance by calculating the cumulative momentum. Finally, the authors evaluated the generalization performance of the model and found that the model showed good generalization ability under different training set sizes and different feature conditions. The results of the study showed that the authors' proposed model could effectively evaluate players' performance in tennis matches and have strong generalization ability.