Multi Source Heterogeneous Data Confidence Fusion and Spatio Temporal Graph Network for Mental Health Monitoring

Mental health monitoring has become an increasingly critical research area as psychological disorders continue to rise, especially among university students. However, effectively capturing dynamic mental health status using heterogeneous data remains a significant challenge due to data noise, incompleteness, and the lack of unified modeling frameworks. To address this issue, this paper proposes a novel framework that integrates a confidence-based multi-source data fusion mechanism with a spatio-temporal graph neural network (ST-GNN) to model complex psychological dynamics. Specifically, multi-modal data, including speech, physiological, and behavioral signals, are dynamically fused based on modality-specific confidence scores to mitigate the impact of noisy and missing data. Furthermore, a spatio-temporal graph network is constructed to capture both the temporal evolution and inter-individual correlations of mental health states. Extensive experiments on a dataset collected from 300 university students over two academic semesters demonstrate that the proposed method outperforms traditional deep learning and graph-based models in both classification accuracy and psychological score prediction.