Comparative Study of Physiological Signals from Empatica E4 Wristband for Stress Classification

The injurious effects of mental stress on the human body and mind are well known. Many researchers have focused on developing stress monitoring systems using physiological signals obtained from the body to alleviate stress. This study aims to provide a comparative analysis of four physiological signals – Electrodermal Activity (EDA), Heart Rate (HR), Skin Temperature (SKT), and Blood Volume Pulse (BVP), recorded using the Empatica E4 Wristband, in building stress classification models. We collect a dataset on 21 participants comprising their physiological signals while they perform a mental arithmetic task, which acts as a stress inducer. We compare the classification accuracy of machine learning classifiers trained on feature sets built using the four signals taken one at a time, two at a time, three at a time and all together. We achieve the highest accuracy of 99.92% using all the four signals. When we consider three signals at a time, EDA, HR, and BVP feature set achieves the best accuracy of 99.88%, and taking two signals at a time, EDA and HR feature set obtains the best accuracy of 99.09%. This paper can act as a guide for a manufacturer to select an optimal set of physiological signals for building efficient stress monitoring systems.