EEG-based stress classification using time-domain features and segmentation techniques.

Stress has been recognized as a significant global health issue, affecting the majority of the population. Rapid and accurate detection of stress is critical for stress treatment. A considerable part of prior work has focused on classifying electroencephalography signals to enable preliminary detection of stress. Previous work has demonstrated considerable success in stress classification/detection using electroencephalography signals. The proposed work aims to classify human stress using electroencephalography signals to enable early intervention. In this research, we have worked with a dataset comprising nearly 211 individuals and proposed a method based on time-domain analysis. Segmentation techniques are used to analyze stress from EEG signals. Both overlapping and non-overlapping methods are employed in this research work. The EEG signals last approximately 480 seconds. We have used the Perceived Stress Questionnaire (PSQ) for the labeled classes of 'stressed' and 'non-stressed'. Different classifiers have been employed to distinguish between stressed and non-stressed classes. Our proposed method achieved an accuracy of 96.32% using a K-nearest neighbors classifier with a non-overlapping segmentation technique.