Digital Twin-Based AI System for Mental Health Monitoring in Academic Environments

This paper presents a Digital Twin–based Artificial Intelligence (DT–AI) framework for real-time mental health monitoring in academic environments. The system combines multimodal data such as facial expressions, voice tone, physiological signals and behavioral features, all collected by IoT-based sensors. A novel hybrid Swin Transformer–Temporal Graph Neural Network (TGNN) architecture is proposed to capture spatial–hierarchical features and temporal dependencies across modalities. The resulting outputs are then fused to compute a Mental Health Index (MHI), which quantifies emotional stability and stress intensity. Each student’s Digital Twin dynamically mirrors real-world psychological states through a continuous flow of data, enabling it for early detection of stress and adaptive interventions. Experimental evaluation on the RAVDESS and WESAD datasets achieved 93.2% emotion-recognition accuracy, 0.88 F1-score for stress classification, and 0.93 MHI correlation, which confirms the effectiveness of this framework. The proposed DT–AI approach provides a scalable, interpretable, and privacy-preserving basis for proactive mental health care in educational ecosystems.