Integrative analysis of psychosocial, chronotype, and environmental predictors of aggressive behavior in Adolescents: Insights from machine learning

Aggressive behavior in adolescents and young adults is a significant public health concern associated with adverse educational, social, and mental health outcomes. This study aimed to identify key predictors of aggression using a cross-sectional dataset from a large, longitudinal U.S. cohort. The outcome was self-reported aggressive behavior, and predictors spanned demographic, psychosocial, behavioral, and contextual domains, including adverse life events, impulsivity, family conflict, peer and school environments, and chronotype.Multiple models were evaluated, including linear regression, a hypertuned random forest, a tuned gradient boosting machine (GBM), XGBoost, and an ensemble model combining random forest and GBM predictions. All models were trained using five-fold cross-validation across five multiply imputed datasets. Linear regression achieved the highest predictive accuracy (r = 0.313; MSE = 40.76), followed closely by the random forest (r = 0.311; MSE = 40.71). The ensemble and GBM models showed slightly lower performance. Across models, key predictors included adverse life events, delayed chronotype, peer network health, family cohesion, and normalized household income.These findings underscore the contribution of environmental and psychological stressors to adolescent aggression, particularly the buffering role of cohesive peer and family relationships. Despite similar predictive accuracy across models, machine learning methods offered advantages for variable importance ranking and interaction discovery. Results highlight the utility of integrating diverse psychosocial, behavioral, and contextual measures to better understand complex behavioral outcomes and inform targeted prevention strategies.