PSYCHOLOGICAL FEATURE AND GENERATIVE AI INTEGRATION FOR EARLY DEPRESSION DETECTION FROM SOCIAL MEDIA: AN INTERPRETABLE COMPUTATIONAL INTELLIGENCE FRAMEWORK

Depression remains a critical global health challenge, demanding scalable, interpretable, and real-time detection systems. This study proposes a novel generative AI framework that integrates clinically validated psychological features with large language model (LLM)-driven text augmentation to enhance depression prediction from social media data. Psychological features were derived from the Beck Depression Inventory (BDI) and embedded into machine learning models to align computational outputs with clinical interpretation. Short and sparse user-generated posts were enriched through GPT-4-based data augmentation. Comparative experiments using Naive Bayes (NB), Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM) classifiers across two public Twitter datasets demonstrated that the proposed approach mitigates feature sparsity, improves classification robustness, and maintains interpretability. The fusion of psychological feature embedding and generative augmentation provides a methodological bridge between computational intelligence and mental health practice, offering a scalable solution for early intervention in population-wide depression monitoring.