Wearable physiological signals under acute stress and exercise conditions

Abstract

In this work, a novel dataset containing physiological signals recorded non invasevely during structured acute stress induction, as well as aerobic and anaerobic exercise sessions is presented. The physiological data were collected using the Empatica E4, a wearable device that measures electrodermal activity, skin temperature, three-axis accelerometry and blood volume pulse, from which heart rate and heart rate variability features can be derived. A stress induction protocol was designed using mathematical and emotional tasks to elicit physiological responses. For aerobic and anaerobic exercise, a stationary bike routine was developed to distinguish between the two types of activity. The dataset includes records from 36 healthy individuals during the stress protocol, 30 during aerobic exercise, and 31 during anaerobic exercise. Several machine learning algorithms were applied to validate the dataset, with XGBoost achieving an accuracy of 93% in classifying stress versus rest, 91% in distinguishing between aerobic and anaerobic exercise, and 84% in a four-label classification task involving stress, rest, aerobic, and anaerobic activities. The dataset is publicly available for further research.