An empirical study on the promoting effects of college sports dance courses on students' mental health

Based on an intelligent algorithm-driven human posture recognition framework, this study constructs a psychological state prediction model integrating skeletal temporal features and movement rhythm characteristics. The model employs ST-GCN and multi-scale temporal convolutions to enhance movement representation. Performance validation on Guangxi university sports dance classroom data yielded a Mean Absolute Error (MAE) of 2.13 for predicting psychological scale scores, outperforming the 2.67 MAE of models without rhythm feature integration. Significant differences emerged between high-engagement (MAE 1.87) and low-engagement (MAE 2.94) student groups, indicating that movement features effectively reflect emotional regulation changes.