Using machine learning to identify the strongest personal, school, and family correlates of youth mental health and well-being.

Research has identified numerous personal, school, and family factors associated with youth mental health, yet it remains unclear which correlates best predict mental health outcomes accurately. This study used machine learning to identify the strongest personal, school, and family correlates of mental well-being, overall mental health problems, and internalizing (anxiety and depression) and externalizing (attention deficit and hyperactivity disorder, conduct disorder, and oppositional defiant disorder) symptoms among school-age children and youth and examines whether associations vary across demographic subgroups. Data were drawn from the 2014-2015 School Mental Health Survey in Canada. Machine learning-based feature importance analyses on 24,692 youth (12,754 [51.7%] girls) identified school belonging as the strongest correlate of general mental well-being, lifestyle behaviors (e.g., sleep and exercise) as the strongest correlate of internalizing symptoms, and class preparedness as the strongest correlate of externalizing symptoms. By leveraging generalized random forest models, we found that the strongest correlates of general well-being and internalizing symptoms, but not externalizing symptoms, exhibited statistically significant group differences by gender, family immigration background, and parent education. For example, although school belonging was the strongest correlate of general mental well-being for both genders, higher belonging was associated with greater well-being among girls compared with boys. In addition, higher academic achievement and family relationship quality were associated with fewer internalizing symptoms among youth from immigrant families than those from nonimmigrant families. These findings highlight unique sets of personal, school, and family factors related to mental health among diverse youth, which may inform tailored interventions targeting these multilevel factors. (PsycInfo Database Record (c) 2026 APA, all rights reserved).