<![CDATA[Adolescent depression is an important mental health concern associated with psychological conditions and excessive digital media use. The extreme imbalance of depression labels in behavioral datasets can reduce the ability of classification models to recognize minority cases. This study aims to develop a TabNet-based deep learning model for classifying adolescent depression labels using social media addiction, stress, anxiety, and related behavioral features. The study used a secondary dataset consisting of 1,200 adolescent samples. Data preprocessing included categorical feature encoding, stratified training and testing data splitting, feature standardization, and the application of the Synthetic Minority Oversampling Technique (SMOTE) to the training data. The TabNet Classifier was trained using Cross-Entropy Loss and the Adam optimizer with a step-decay learning rate and early stopping mechanism. The experimental results showed an accuracy of 99.15%, precision of 98.32%, recall of 100%, F1-score of 99.15%, and ROC AUC of 1.0000, with optimal performance achieved at epoch 53. These findings indicate that TabNet can effectively learn psychological and digital behavioral patterns for adolescent depression label classification. The proposed approach provides a potential computational framework for data-driven mental health risk classification, although further validation using diverse empirical datasets is required.]]>
Copyrights © 2026
