Recreational Activities and Youth Outcomes: An Explainable Machine Learning Study

Recreational activities are considered vital for emotional and cognitive health of youth, yet empirical research examining the specific associations between recreational activities and youth behavioral and cognitive outcomes is limited.

In this cohort study, interpretable machine learning methods, specifically random forest paired with SHapley Additive exPlanations (SHAP) analysis, were applied to quantify the contribution of each activity to behavioral and cognitive outcomes in the Adolescent Brain Cognitive Development℠ (ABCD) Study baseline data (ages 9-10). Data-driven activity groups were defined by SHAP contribution patterns for behavioral and cognitive outcomes. Engagement in these groups was then tested for association with the corresponding outcomes using linear mixed-effects models in the hold-out sample, both cross-sectionally and longitudinally.

Of activities, 14 were identified with more favorable and 8 with less favorable behavioral patterns, and 13 were identified with more favorable and 2 with less favorable cognitive patterns. Activities with more favorable behavioral patterns (eg, skiing/snowboarding) were linked to fewer problems (β = -1.12, p < .001), whereas activities with less favorable patterns (eg, martial arts) were linked to more problems (β = 1.75, p < .001). For cognition, activities with less favorable patterns (eg, football) were associated with lower cognitive scores (β = -1.34, p < .001), whereas activities with more favorable patterns (eg, musical instruments) were associated with higher scores (β = .98, p < .001), and family income moderated this association (p < .001).

Recreational activities exhibit distinct patterns of association with cognitive and mental health outcomes. These data-driven findings highlight the importance of considering individual differences and contextual factors when investigating youth engagement in recreational activities.