Multimodal Stress Classification Using Physiological Signals and Facial Expressions with Machine and Deep Learning Models

Read full text Summarise with VIRAT
Authors: Bipesh Raj Subedi
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Stress detection has become increasingly important in modern healthcare and wellness applications. This paper presents a comparative study of machine and deep learning approaches for stress classification using physiological signals and facial expressions. Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) models are evaluated for analyzing physiological data including electrodermal activity, blood volume pressure, skin temperature, and accelerometer readings across 15, 30, and 60-second time windows. VGG16 and a Custom Convolutional Neural Network (CNN) are deployed for stress classification from 48x48 grayscale facial images data. Using the WorkStress3D dataset, experimental results demonstrate that LSTM achieves test accuracies of 80%, 85%, and 86% for the data with 15, 30, and 60-second time windows respectively, outperforming SVM. For facial expressions, VGG16 attains 87% test accuracy, surpassing the custom CNN's 81%. The findings reveal that longer time windows enhance physiological signal-based predictions, while both CNN architectures effectively capture stress-indicative facial features, establishing a foundation for robust stress monitoring systems.

Fields of Study & Subjects

Emotion and Mood Recognition Heart Rate Variability and Autonomic Control Context-Aware Activity Recognition Systems

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN
Document type
Preprint