Stress Recognition - A Step Outside the Lab

Abstract

Despite the potential for stress and emotion recognition outside the lab environment, very little work has been reported that is feasible for use in the real world and much less for activities involving physical activity. In this work, we move a step forward towards a stress recognition system that works on a close to real world data set and shows a significant improvement over classification only systems. Our method uses clustering to separate the data into physical exertion levels and later performs stress classification over the discovered clusters. We validate our approach on a physiological stress dataset from 20 participants who performed 3 different activities of varying intensity under 3 different types of stimuli intended to cause stress. The results show an f-measure improvement of 130% compared to using classification only. Copyright © 2014 SCITEPRESS - Science and Technology Publications. All rights reserved.