PAMG-AT: Graph Neural Network for Wearable Stress Detection

PAMG-AT (Physiological Attention Multi-Graph with Adaptive Topology) is a graph neural network for stress detection from wearable physiological signals. Each sensor channel (ECG, EDA, EMG, respiration, temperature, accelerometer) becomes a node in a learned, adaptive graph so the model can discover which signals matter for each subject and condition. On the WESAD benchmark it reaches 94.59% accuracy with chest-only sensors and 92.80% with a hybrid chest-plus-wrist setup under leave-one-subject-out cross-validation, which is the strict protocol that generalises to unseen people. Manuscript under review at Biomedical Signal Processing and Control (2026), co-authored with Prof. Abdulhamit Subasi at the University at Albany. Stack: Python, PyTorch Geometric, neurokit2, SLURM on a DGX A100 cluster.