Academic pressures impact college students' mental health, emphasizing the need for timely stress detection. Traditional assessment methods are often subjective and insufficient for ongoing monitoring, unlike current AI approaches. This research proposes an AI-driven early warning system for mental health risk prediction using multimodal Industrial Internet of Things (IIoT) data, EEG, HRV, behavioral indicators, and interaction logs. The system employs an Adaptive Manta-Ray Foraging Optimized Graph Neural Network (AMFO-GNN) utilizing multimodal data and graph-based reasoning enables adaptive, and interpretable real-time predictions of mental health risk patterns through the capture of complex inter-student and intra-feature interactions. Experiments on a campus dataset, 8076 samples collected from College Student Mental Health scenarios, demonstrate robust performance. The model predicts mental health risks such as stress, anxiety, and depression among college students, providing clinically interpretable outputs for early intervention. Data preprocessing includes noise filtering, normalization, and temporal alignment to ensure high-quality inputs. Feature extraction includes EEG analysis using wavelet transforms, HRV features from statistical and spectral analysis, facial expressions and gestures via CNN embeddings, and text/clickstream logs encoded with transformer-based embeddings. Extracted features are integrated using a tensor fusion network (TFN), enabling effective cross-modal interaction and comprehensive representation. AMFO optimizes model parameters through an adaptive multi-objective feature optimization technique. GNN learns relational dependencies via structured graph-based node feature aggregation. The system achieves high stress classification accuracy (0.951) outperforming traditional models implemented using Python. The proposed approach highlights the integration for intelligent mental health monitoring and adaptive human–computer interaction.
Yan et al. (Sat,) studied this question.