Using explainable machine learning to classify subjective wellbeing status in Chilean adolescents

Adolescent subjective well-being (SWB) is a key indicator of psychosocial adjustment, yet its correlates are multifactorial and difficult to model using traditional linear approaches. The aim of this study was to develop and evaluate explainable supervised machine-learning models to classify low versus high SWB in Chilean adolescents using psychological, behavioural, and sociodemographic factors, and to interpret model predictions to characterize multivariate predictive risk and protective profiles. A cross-sectional sample of 913 students (10 − 19 years) from public schools in the Biobío region completed measures of emotional self-regulation, depressive symptom indicators, health-related habits, leisure-time motivation, and sociodemographic characteristics. SWB was assessed with the Personal Wellbeing Index – School Children (PWI-SC) and operationalized as high (PWI-SC ≥ 7) versus low (PWI-SC < 7) to support an interpretable binary classification framework. Among the tested algorithms, XGBoost achieved the best discrimination on the held-out test set (AUC = 0.86). SHAP analyses revealed that depressive symptoms and long-term emotional regulation contributed most strongly to the model´s predictions, followed by healthy eating, physical activity, and motivational dimensions. Overall, explainable machine-learning models can help describe multivariate predictive patterns of adolescent SWB in public-school settings. These findings must be interpreted as predictive associations rather than causal effects.