Predicting athletic performance in track and field athletes based on wearable physiological and psychological indicators: an interpretable machine learning study.

Track-and-field performance fluctuates week to week under the joint influence of training load, recovery, autonomic regulation, and psychological readiness, and wearables now enable continuous physiological monitoring; however, whether bodily signals and readiness markers contribute differently to performance prediction across technical and physical events remains unclear. In a prospective longitudinal study, 132 competitive track-and-field athletes (71 physical-event, 61 technical-event) were monitored over a planned 14-week athlete-week design. One-week lagged morning LnRMSSD, resting heart rate, sleep duration and quality, session-RPE load, cognitive anxiety, and self-efficacy were used to predict subsequent standardized weekly performance change. Extreme gradient boosting (XGBoost) models were evaluated with athlete-level grouped cross-validation. SHapley Additive exPlanations (SHAP) summarized feature contributions, and OLS interaction models with athlete-clustered robust standard errors tested moderation by event type. The integrated XGBoost model showed the highest overall prediction among the evaluated feature sets (mean outer-fold R² = 0.510, RMSE = 0.505, MAE = 0.406; within-athlete centered sensitivity R² = 0.177-0.306), and readiness measures added a consistent increment beyond monitoring-only features (mean ΔR² = 0.088). SHAP identified the ordinal competitive-level code as the largest single contribution (23.4%), followed by lagged physiological and self-efficacy features. Interaction analyses gave partial evidence of event-sensitive moderation, mainly for psychological readiness in technical events, whereas aggregated physiological and load interactions were non-significant. Sensitivity analyses showed that predictive performance decreased after removing the competitive-level code (R² = 0.366) and in within-athlete centered models, indicating that the headline R² reflected both stable athlete-level stratification and week-to-week monitoring signals. Overall, integrated wearable physiological and psychological monitoring can help predict and characterize week-to-week performance change in track-and-field athletes, with the most consistent event-sensitive signal involving psychological readiness, particularly self-efficacy, in technical events, while not implying causal mechanisms.