KMAN Publication Inc. (KMANPUB) Journal of Psychosociological Research in Family and Culture 3041-8550 In Press 2026 07 01 Explainable Machine Learning Prediction of Adolescent Depression From Family Functioning, Parenting Practices, Peer Relationships, Cultural Values, and Digital Media Behaviors Explainable Machine Learning Prediction of Adolescent Depression From Family Functioning, Parenting Practices, Peer Relationships, Cultural Values, and Digital Media Behaviors 1 21 EN Journal Article 2025 12 30 <table> <tbody> <tr> <td> <p><strong>Objective: </strong>This study aimed to develop, validate, and explain a machine-learning model for predicting elevated depressive symptoms among adolescents using family functioning, parenting practices, peer relationships, cultural values, and digital-media behaviors.</p> <p><strong>Methods and Materials:</strong> A cross-sectional predictive study was conducted with 1,000 Canadian adolescents aged 13–18 years who were recruited through stratified multistage cluster sampling. Depression was assessed using the Patient Health Questionnaire for Adolescents, while validated instruments measured family functioning, parenting practices, peer attachment, cultural-value conflict, problematic social-media use, and related behavioral factors. The dataset was divided into development and independent holdout samples. Elastic-net logistic regression, random forest, gradient-boosted decision trees, extreme gradient boosting, support-vector machines, and multilayer perceptron models were compared. Model performance was evaluated using ROC-AUC, precision-recall AUC, sensitivity, specificity, balanced accuracy, F1 score, and Brier score. SHapley Additive exPlanations were used to determine global feature importance, directional effects, nonlinear thresholds, and predictor interactions.</p> <p><strong>Findings:</strong> Extreme gradient boosting demonstrated the strongest predictive performance, achieving a ROC-AUC of .879, precision-recall AUC of .756, sensitivity of .842, specificity of .762, balanced accuracy of .802, F1 score of .723, and Brier score of .132. It outperformed elastic-net logistic regression and all other candidate algorithms. Poor family functioning was the most influential predictor, followed by peer alienation, problematic social-media use, nighttime device use, cybervictimization, inconsistent discipline, and cultural-value conflict. Positive parenting, longer sleep duration, trusted-friend availability, and supportive online-community engagement reduced predicted risk. Important interactions were observed between family dysfunction and peer alienation, problematic social-media use and nighttime device use, and cybervictimization and trusted-friend availability. Model performance remained stable across alternative depression cutoffs and major demographic subgroups.</p> <p><strong>Conclusion:</strong> Explainable machine learning can identify adolescent depression risk with good discrimination and calibration by integrating relational, cultural, and digital-behavioral factors, while SHAP analysis provides clinically meaningful explanations of individual and global predictions.</p> </td> </tr> </tbody> </table> https://www.journals.kmanpub.com/index.php/jprfc/article/download/6262/11724