Associations between physical activity and adolescent well-being: a loneliness-moderated chain mediation model with explainable machine learning.

Introduction: This cross-sectional survey examined how physical activity was associated with adolescent well-being through cognitive reappraisal and psychological resilience, and whether these associations varied by loneliness. Methods: Data from 964 adolescents aged 10-19 years were analyzed using moderated chain mediation models. XGBoost regression and SHAP interpretation were used as supplementary predictive tools. Results: Physical activity showed no significant direct association with well-being after cognitive reappraisal and psychological resilience were included. Significant indirect associations were observed through cognitive reappraisal, through psychological resilience, and through the sequential pathway linking cognitive reappraisal with psychological resilience. Loneliness moderated the association between psychological resilience and well-being, with a stronger positive association among adolescents reporting higher loneliness. The XGBoost model showed acceptable internal predictive performance in the test set (RMSE = 2.599, MAE = 1.860, R² = 0.640). SHAP analysis indicated that psychological resilience made the largest predictive contribution, followed by loneliness and cognitive reappraisal, whereas physical activity made a smaller contribution. Partial dependence results suggested a modest nonlinear pattern for physical activity, with predicted well-being increasing mainly from low to moderate activity levels and then tending to plateau. Discussion: Overall, the findings suggest that physical activity is linked to adolescent well-being mainly through psychological resources. The observed associations should be interpreted as cross-sectional and predictive patterns rather than causal effects.