Automatic and Intelligent Stressor Identification Based on Photoplethysmography Analysis

Abstract

A stressor is an external stimulus that an individual might consider demanding or threatening regarding his/her safety. Stressors cause the release of stress hormones and can be categorized into two categories: physiological and psychological. Wearable technology has been on the rise in the past decade, and the advancement of IoT will only further the benefits that wearable technology will bring. This work leverages the output of wearable technology to provide automatic stress and stressor identification model. In particular, this study proposes a novel algorithm that first detects instances of stress and then classifies the stressor type using photoplethysmography (PPG) data from wearable smartwatches. The time-series PPG data is transformed into 2D spatial images by first extracting the interbeat interval (IBI), and the blood volume pulse (BVP) features from the PPG, then transforming these data to 2D image data. The developed model successfully identified stress instances using IBI-BVP spatial domain images with an average accuracy of 98.10% with a convolutional neural network (CNN) and 99.18% using the average pixel intensity of these images with the extra trees classifier. It also successfully categorized stressor types into physical, cognitive, and social with an accuracy of 98.5% using the CNN and 96.45% using the extra trees classifier, outperforming state-of-the-art.