Integrating Graph Neural Networks and Temporal Graph Convolutions for Enhanced Multimodal Stress Detection in Physiological and Behavioral Data Streams

Stress has a substantial impact on mental and physical health, which highlights the necessity of reliable detection systems that make use of multimodal data. In order to assess stress patterns, this study makes use of the Stress-Lysis dataset, which contains characteristics like step count, temperature, humidity, and stress levels. To identify temporal trends and interdependencies in the data, a Graph Neural Network (GNN) combined with Temporal Graph Convolutions (T-GCN) is suggested. With an accuracy of 92.3% and a precision of 91.7%, the suggested GNN-T-GCN model outperformed five baseline methods. These results show how the model can forecast stress in real time, opening the door to more sophisticated wearable stress monitoring.