A Deep Learning-Based Study of Player Styles and Cross-League Performance Adaptation Mechanisms: A Case Study of the NBA and CBA

云瀚 肖¹伟平 李^2*建刚 陈¹

• ¹西安体育学院运动与健康科学学院, 西安市, China
• ²中国体育科学学会体育统计分会, 西安市, China

This study proposes an approach that combines deep learning with interpretable analysis to investigate player styles and the mechanisms underlying cross-league performance, with the goal of quantitatively revealing how stylistic features influence athletes' performance across leagues. Using game data from players and teams during the 2019-2024 seasons as samples, the study clusters and models players' technical styles via PCA, t-SNE, and Gaussian mixture models, and applies a branch-type multilayer perceptron (Branch-MLP) together with the SHAP algorithm to perform an interpretable analysis of mainstream tactical structures. The results show that the NBA places greater emphasis on offensive efficiency and team coordination, whereas the CBA focuses more on ballpossession control and physical confrontation. The Branch-MLP achieves higher accuracy in tactical recognition. Further quantitative analysis indicates that interior defense-oriented role players exhibit more stable cross-league performance, while perimeter ball-handling role players experience greater fluctuations. This study enriches quantitative methods for analyzing athletic performance and offers reliable data references for player training, transfer evaluation, and youth development strategies.