KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 50 2026 10 10 Predicting Adolescent Depression From Digital Behavior, Sleep, and Social Connectedness Using CatBoost With SHAP Explainability Predicting Adolescent Depression From Digital Behavior, Sleep, and Social Connectedness Using CatBoost With SHAP Explainability 1 18 10.61838/kman.jayps.6304 EN Journal Article 2026 07 19 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> This study aimed to predict depressive symptom severity among Canadian adolescents from digital behavior, sleep characteristics, and social connectedness using CatBoost and to explain the contribution of individual predictors through SHAP analysis.</p> <p><strong>Methods and Materials:</strong> A cross-sectional predictive study was conducted among 1,238 Canadian adolescents aged 13–18 years. Depressive symptoms were assessed using the Patient Health Questionnaire-9 modified for adolescents (PHQ-9A). Digital behavior variables included recreational screen time, smartphone use, social media use, checking frequency, nighttime device use, and number of social media platforms. Sleep was assessed using the Pittsburgh Sleep Quality Index and supplementary measures of sleep duration, latency, and bedtime variability. Social connectedness was measured using the Social Connectedness Scale–Revised, with additional indicators of family support, peer support, loneliness, and face-to-face interaction. The dataset was divided into 80% training and 20% independent test samples. CatBoost regression was tuned using five-fold cross-validation and compared with random forest, conventional gradient boosting, elastic net, and multiple linear regression. Model interpretability was examined using SHapley Additive exPlanations.</p> <p><strong>Findings:</strong> The CatBoost model achieved MAE = 2.72, RMSE = 3.57, and R² = 0.61 in the independent test sample, outperforming random forest (R² = 0.53), gradient boosting (R² = 0.56), elastic net (R² = 0.41), and multiple linear regression (R² = 0.38). SHAP analysis identified poor sleep quality as the strongest predictor, followed by lower social connectedness, greater loneliness, more frequent nighttime device use, shorter sleep duration, greater recreational screen time, and higher bedtime variability. Higher family support, peer support, face-to-face interaction, and social connectedness were associated with lower predicted depressive symptom severity.</p> <p><strong>Conclusion:</strong> Adolescent depressive symptoms were predicted with substantial accuracy from a multidimensional combination of sleep, digital behavior, and social connectedness, with sleep quality and interpersonal factors contributing more strongly than overall technology exposure.</p> </td> </tr> </tbody> </table> https://journals.kmanpub.com/index.php/jayps/article/download/6304/12105