Stress and anxiety detection: deep learning and higher order statistic approach

Abstract

Today's teenagers are dealing with anxiety and stress. Anxiety, depression, and suicide rates have increased in recent years because of increased social rivalry. The research is focused on detecting anxiety in students due to exam pressure to reduce the potential harm to a person's wellness. Research is performed on databases for anxious states based on psychological stimulation (DASPS) and our own database. The measured signal is divided into sub bands that correspond to the electroencephalogram (EEG) rhythms using the Butterworth sixth-order order filter. In higher dimensional space, the nonlinearities of each sub-band signal are analyzed using higher order statistics third-order cumulants (TOC). We have classified stress and anxiety using the support vector machine (SVM), K-nearest neighbor (K-NN), and deep learning bidirectional long short-term memory (BiLSTM) network. In comparison to previous techniques, the proposed system's performance using BiLSTM is quite good. The best accuracy in this analysis was 87% on the DASPS database and 98% on the own database. Finally, subjects with high stress levels had more gamma activity than subjects with little stress. This could be an important attribute in the classification of stress.