Digital Behavior and Mental Health Prediction Through Explainable AI

Digital habits such as screen time, notifications, and social media engagement have increasingly influenced mental health and overall well-being. This research examines the link between these digital practices and mental health outcomes by utilizing an explainable AI framework from a public dataset containing 500 anonymized entries that combine behavioral metrics with self-reported measures. Building on initial logistic regression analyses, this study employs gradient boosting with XGBoost, enhanced by Shapley Additive Explanations (SHAP), to strengthen both predictive accuracy and interpretability. To evaluate reproducibility, models were trained with five random seeds, and performance was assessed using root mean square error (RMSE) and the coefficient of determination (R²). The outcomes demonstrated consistent predictive performance (RMSE ranged from 5.8 to 6.8; R² ranged from 0.25 to 0.31) and consistently highlighted sleep hours, notification count, and focus score as the most significant predictors. SHAP analysis revealed low variance across seeds, reaffirming the reliability of these features. These findings highlight how behavioral data can inform digital wellness initiatives. This research contributes to a transparent, reproducible analytical framework that bridges the gap between computational modelling and psychological research, supporting the application of explainable AI in mental health research.