A deep learning approach for assessing stress levels in patients using electroencephalogram signals

Abstract

Stress can contribute to many health problems, such as high blood pressure, heart disease, obesity, and diabetes. Therefore, early stress detection is essential in preventing illness and health problems. This study proposes a music experimentation approach to identify stress levels among the subjects. In this experiment, we study 45 subjects in the age category of 13–21. The model architecture used in the study is EEGnet, a compact convolutional neural network with a Relu activation function. We experimented with the mother wavelet decomposition method with 0 to 60 Hz frequency electroencephalogram signals and frequency division in 5 bands. The mounting positions involved are frontal, temporal, partial, and central. Signals generated at the frontal and temporal position in band value of 8–16 Hz serve as the most prominent feature in an experiment. We have achieved an accuracy of 99.45% for the alpha band with deep learning EEGnet architecture