Structured environmental cues and youth sports consumption: an AI-powered behavioral modeling approach.

Introduction: Physical inactivity among children and adolescents represents a critical global public health challenge, imposing an annual economic burden exceeding $27 billion on healthcare systems worldwide. Despite this pressing concern, the complex, cross-level relationships between structured environmental cues-spanning household, school, and community domains-and youth sports consumption remain inadequately understood, particularly across the heterogeneous urbanrural continuum characteristic of rapidly developing economies. Methods: This study employed a longitudinal survey (China Health and Nutrition Survey, CHNS) comprising 4,240 child-wave observations nested within 940 households and 182 communities. We benchmarked conventional multilevel regression against three machine learning approaches-gradient boosting, random forest, and LSTM sequence models-using temporal and spatial holdout validation. Feature-level and interaction-level interpretability was recovered via SHAP decomposition to quantify the magnitude and breakdown of urbanrural interplay in multi-layered environmental factors shaping youth sport consumption. Results: The gradient boosting model significantly outperformed the multilevel baseline, achieving a holdout Discussion: This study provides novel evidence that public space recreation infrastructure yields substantially greater behavioral returns in resource-limited communities, challenging assumptions that uniform investment strategies are optimal across socioeconomic strata. The findings offer a new methodological and intervention paradigm for promoting youth physical activity in socioeconomically diverse, rapidly urbanizing regions, emphasizing targeted infrastructure investment in disadvantaged communities as a high-return public health strategy.