Impact of Adaptive Energy Dependable and Statistical Power Variation Features in the Classification of Cognitive Stress

The paper proposes an adaptive energy-dependable feature extraction approach for classifying mental stress. The electroencephalogram (EEG) signal components are initially separated into lower and higher frequency components using two bandpass filters. The peak points are then detected on the lower frequency components that provide the higher amplitude components with the neighborhood samples. The energy and slope between the adjacent peak points are also used in estimating the adaptive energy-dependent features. Along with the adaptive features, statistical power variation features are also estimated based on the power variation between the lower and higher frequency bands. The proposed Stress classification approach results in an average precision, specificity, and accuracy of 95.89%, 99.61%, and 96.86% respectively when evaluated using the SAM-40 dataset which is higher than the traditional stress classification approaches. The approach classifies the stress categories Stroop stress, arithmetic task stress, and mirror image recognition task stress with three stress intense levels.