Approximate Model for Stress Assessment Using Electroencephalogram Signal

Abstract

Mental stress is a problem that people may often face. Although there are some psychobiological stress measurement methods based on electroencephalogram (EEG) signals, these methods use expensive medical equipment to gather multichannel signals and cannot measure stress in real time in daily life. Many novel wearable devices now have a sensor for physiological signal detection, which helps people manage health conditions. Several studies have recently investigated the application of wearable devices to stress detection. In this paper, an approximate model based on an EEG signal is designed for measuring stress. To establish the connection between the EEG and electrocardiogram signals, we use wearable devices to simultaneously collect the two types of data from volunteers. The exponentially weighted moving average is used to smooth out the EEG power spectrum features (α, β, etc.). An EEG-based feature vector is constructed to predict stress scores, with polynomial regression used to build the model. The experimental results show that the proposed method achieves a symmetric mean absolute percentage error of 17.44 and a root mean square error of 11.26.