A Deep Learning Approach Using WESAD Data for Multi-Class Classification with Wearable Sensors

Abstract

Stress is a common part of life, but chronic or intense stress can severely impact safety and disrupt daily activities. Early detection of mental stress can prevent many related health issues. Stress causes significant changes in biosignals, which can be used to identify stress levels. This paper proposes various machine learning and deep learning techniques for detecting stress using multimodal datasets from wearable physiological and motion sensors, aiming to prevent stress-related health problems. Data from sensor modalities such as three-axis acceleration (ACC), electrocardiogram (ECG), blood volume pulse (BVP), body temperature (TEMP), respiration (RESP), electromyogram (EMG), and electrodermal activity (EDA) were taken from the WESAD dataset, covering amusement, neutral, and stress states. The best results for the Adam optimizer were achieved with a batch size of 16, yielding an accuracy of 90.26%, while the best results for the SGD optimizer were obtained with a batch size of 16, yielding an accuracy of 87.84%.