Abstract:

Abstract Mental stress is proving to be a cause for functional impairment of daily activities and it is on increase. Further, continual stress should implicate numerous disorders of mind and body. Stress increases the chances of despair, stroke, coronary failure, and cardiopulmonary arrest. Human brain is a major target of psychological pressure because it determines the context of the human mind in a threatening and demanding circumstances as shown by latest neuroscience. The objective method of determining the level of stress, taking into account the human brain, greatly increases the associated dangerous effects. Therefore, the system proposed in this paper performs electroencephalography (EEG) signal analysis. Data for stressed individuals is recorded and the signal is filtered with time domain and frequency domain-based filters. Fast Fourier Transform (FFT) algorithm is used to transform the data from time domain to frequency domain. Features namely Normalized Absolute Power, Relative Power, Normalized Peak Power and Change in Power are extracted and paired t-test is used for feature selection. Features having confidence value above 95% are chosen. Within the experimental setting, stress is induced through Mental Arithmetic Task Tool (MATT) which is popular experimental pattern found on the concept of Montreal Imaging Stress Test (MIST). When performance was evaluated of all the subjects it was observed that in normal condition the average performance is 73.71% and in stress condition it is 60.18%. So, it is evident that MATT is inducing stress as performance is reduced by average 13.53% from normal to stress. The proposed system involves EEG feature extraction and feature selection using paired t-test to various brain locations across six frequency bands for stress detection. In this paper, our aim is to compare the different types of feature values of appropriate channels and frequency band to find the confidence percentage (above threshold percentage) of feature values that helps to differentiate the normal and stress classes. The results of proposed system find the correct frequency band of appropriate channel of feature types that differentiate the normal and stress data.

Keywords: