Explainable graph attention network for stress recognition (StressGAT) via differential action units

2026-07-22

AI Summary

Existing approaches struggle to capture highly individualized facial expression dynamics and lack clinical interpretability. This work proposes a novel framework integrating graph attention networks with differential Action Units (ΔAUs): ΔAUs normalize expressions relative to each individual’s neutral baseline to model personalized affective changes, while a graph attention mechanism captures dynamic dependencies among Action Units. Combined with multiple instance learning, the framework enables precise localization of peak stress intervals. Evaluated on a stress-induced dataset from 58 participants using leave-one-subject-out cross-validation, the method achieves a subject-independent recognition accuracy of 88.62%. This study is the first to simultaneously achieve high accuracy and interpretability in stress recognition, revealing distinct individual expressive phenotypes.

Abstract

Stress is a dynamic process characterized by significant individual variability in facial expression. Traditional architectures, such as Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs), often overlook person-specific baselines or lack the representational capacity to model the non-linear temporal progression of distress due to sequential bottlenecks and rigid grid-based constraints. Furthermore, many deep learning models lack the interpretability required for clinical deployment. This study introduces StressGAT, a Graph Attention Network that leverages the relational inductive bias of graph modeling to capture complex facial dynamics that indicate acute stress. By using Differential Action Units, the framework normalizes individual responses relative to neutral baselines to achieve personalized recognition. The proposed model achieves 88.62\% accuracy on a diverse stress-induction cohort (58 participants) using a subject-independent, Leave-One-Subject-Out (LOSO) cross-validation protocol. Beyond predictive accuracy, the architecture integrates a Multiple Instance Learning (MIL) attention mechanism to identify peak stress intervals and reveal distinct expressivity phenotypes. By simultaneously optimizing for accuracy and interpretability, this framework provides a robust, explainable solution for personalized affective monitoring.

Problem

• stress recognition
• individual variability
• facial dynamics
• interpretability
• personalized affective monitoring

Innovation

• Graph Attention Network
• Differential Action Units
• Explainable AI
• Personalized Stress Recognition
• Multiple Instance Learning