Psychological State Analysis of Swimmers Based on Machine Learning and Multi-dimensional Data

This study utilizes longitudinal training and physiological data from 10 athletes of the Shanghai Youth Swimming Team (January–June 2024) to construct a predictive model for psychological states. Heterogeneous daily training logs and physiological records were systematically vectorized into a multi-dimensional dataset (\(N=599\)) comprising 757 features. By comparing 12 machine learning classification algorithms—including Random Forest, XGBoost, and LightGBM—via five-fold stratified cross-validation, we identified the BaggingClassifier as the optimal model. It achieved 75% accuracy on an independent test set, effectively mapping complex physiological and training variables to psychological assessment outcomes. The proposed framework provides a data-driven, quantifiable decision support tool for monitoring and enhancing the mental well-being of youth athletes.