Search papers, labs, and topics across Lattice.
This study introduces StressGAT, an Explainable Graph Attention Network designed for personalized stress recognition through the analysis of facial dynamics using Differential Action Units. By overcoming the limitations of traditional deep learning models, such as RNNs and CNNs, StressGAT captures individual variability in stress responses while achieving an impressive 88.62% accuracy in a diverse cohort of 58 participants. The model not only enhances predictive performance but also incorporates a Multiple Instance Learning attention mechanism to identify critical stress intervals, providing valuable interpretability for clinical applications.
StressGAT achieves 88.62% accuracy in stress recognition by personalizing facial expression analysis, all while maintaining interpretability for clinical use.
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.
