<![CDATA[Depression is a mental health problem with high prevalence that requires accurate and reliable computational-based prediction systems to support early detection. This study proposes a depression risk prediction architecture based on a stacking ensemble approach incorporating an out-of-fold (OOF) mechanism to prevent data leakage during meta-feature generation. The model combines Support Vector Machine and XGBoost as base learners, with Logistic Regression employed as the meta-learner. A public Depression Professional Dataset is processed using a stratified split strategy, class balancing on the training data through SMOTE, and feature standardization to enhance training stability. Experimental results demonstrate that the proposed approach achieves superior performance with an accuracy of 0.99, precision of 0.91, recall of 1.00, and an F1-score of 0.95, along with consistent detection capability for the minority class. These findings confirm that the systematic integration of OOF stacking and SMOTE improves model sensitivity while reducing false negative errors, making it suitable for the development of artificial intelligence–based mental health screening systems.]]>
Copyrights © 2026
