EEG-based stress identification and classification using deep learning

Mental stress has become one of the major reasons for the failure of students or their poor performance in the traditional limited-duration examination system. In this study, we aim to find the relationship between the student's level of stress and the deterioration of their subsequent examination results. Furthermore, we want to explore if different EEG frequency bands can be used as biomarkers of stress levels. We collected EEG data from the students while they performed the Montreal Imaging Stress Task-based Mental Arithmetic Tasks (MAT). They performed tests of the same level of difficulty twice; once without any limitation of time (and/or feedback) and next under limited time followed by feedback to induce more stress. We observed that the average score of 95% in the untimed test was dropped to 78% in the case of a timed test and a substantial difference in spectral powers of beta, alpha, and theta frequency bands of EEG. We took this limitation of time as a stressor and comprised three classes based on three stress levels (relaxed during rest, low during an untimed test, and high during a timed test). We used two different deep learning frameworks to classify this data and an accuracy of 70.67% was achieved using Long Short-Term Memory (LSTM) and 90.46% with convolutional neural networks (CNN). We found that time limitation increases the stress level of students and impairs performance, while EEG frequency bands converted to brainwave images, serve as potential biomarkers for stress detection.