Front Psychiatry. 2026 Aug 14;17:1908580. doi: 10.3389/fpsyt.2026.1908580. eCollection 2026.
ABSTRACT
BACKGROUND: Exercise intervention has been associated with cognitive development in children. However, current study methods have not clarified the potential nonlinear association between exercise dosage and cognitive ability, and interpretable machine learning approaches may provide a novel perspective for exploring these complex relationships. Therefore, this study aimed to construct and verify a machine learning model of the relationship between exercise dosage and cognitive development in children.
METHODS: A total of 8623 valid samples from the China Family Panel Studies (CFPS) database were analyzed in this study. SPSS 25.0 was used to perform descriptive statistical analysis. SHAP values and partial dependence plots were used to enhance model interpretability. Three predictive models-Exercise Dose-Cognitive Ability Score (ED-CAS) model, Daily Physical Activity Duration-Cognitive Ability Score (DPAD-CAS) model, and Weekly Frequency of Physical Activity-Cognitive Ability Score (WFPA-CAS) model-were developed using the random forest algorithm.
RESULTS: The ED-CAS, DPAD-CAS, and WFPA-CAS models demonstrated moderate predictive performance (R² = 0.21, 0.14, and 0.11, respectively). SHAP analysis revealed mean marginal contributions of 0.12, 0.07, and 0.11 for ED, DPAD, and WFPA, respectively. Partial dependence analyses further identified normalized parameter peaks at 1.35 (ED), 1.4 (DPAD), and 0.9 (WFPA), indicating that these exercise-related features contributed differently to model predictions across dose ranges. However, dose thresholds were observed (≤720 minutes/week, ≤6 days/week, ≤85 minutes/day), beyond which CAS declined significantly.
CONCLUSIONS: This study identified a potential inverted U-shaped association between exercise dosage and cognitive ability in children. The observed dose ranges may provide preliminary reference information for future exercise intervention. However, longitudinal and experimental studies are required to determine causal relationships and establish evidence-based exercise recommendations.
PMID:42666282 | PMC:PMC13522174 | DOI:10.3389/fpsyt.2026.1908580
