In this study, we first collected and preprocessed data from 500 basketball players between the ages of 13-35 years old in the same-court rivalry training in Northwest China, after which we utilized the Global Chaos Bat Algorithm (GCBA) for the mental training anxiety emotion feature extraction, and analyzed the correlation between each feature and the anxiety emotion through the Pearson coefficient. Finally, the LightGBM-based emotion prediction model was constructed, and the SHAP value was introduced to evaluate the feature importance of the model. The results show that the LightGBM model performs better and has higher prediction accuracy, which is as high as 96.68%; the interpretation results of the SHAP algorithm indicate that the gender and age of the basketball players are the main real-world factors for assessing their anxiety in same-court rivalry training. In addition, their game scores, opponents’ strengths and injury histories during the same-court rivalry training were the main intrinsic factors for their anxiety. In conclusion, the psychological state of basketball players can reflect the severity of their training anxiety, and it further reveals the relationship between the psychological characteristics of basketball players and their training anxiety.