AUTOMATIC STRESS CLASSIFICATION USING PHOTOPLETHYSMOGRAM-BASED IMAGES

Abstract

Stress is a body reaction to any threatening stimulus (stressor) that is perceived. Stress is subjective and is manifested differently from one person to another. Previous literature approaches focus on developing general stress detection models to detect stress status or stress level using different physiological signals obtained by wet gel electrodes and lead wires systems. The rapid growth in the wearable health monitoring technology market makes embedded Photoplethysmography (PPG) sensors in the wearable sensing devices a potential solution that can measure the volumetric variations of blood circulation influenced by the cardiac, vascular, and autonomic nervous systems, all affected by stress. In this research, a new PPG-based image analysis is proposed in which the inter-beat intervals (IBI) and the blood volume pulse (BVP) are extracted from wearable-based PPG signals and converted to spatial domain images and transformed to the frequency domain. These generated images and the calculated average pixel intensity values from the spatial domain images are used to train, validate, and test deep learning and machine learning classification models to classify the individual's stress status into stressed and non-stressed, and the stressor type into physical, cognitive, and social. The developed classification models were evaluated on two datasets, and the results show that our proposed methods successfully classify the stressor type and the participants' stress state with higher accuracy and outperformed state of the art.

Date of Award	2021
Original language	American English
Awarding Institution	HBKU College of Science and Engineering