Classification of Mental Stress using Dry EEG Electrodes and Machine Learning

Stress is a major health issue that affects people worldwide and results in several diseases and negative psychological consequences. Therefore, early detection of stress has become crucial for maintaining a healthy society. In this study, five different machine learning classifiers were studied for their accuracy in assessing mental stress among university students. Mental stress was obtained from EEG signals using a dry Electroencephalography (EEG) electrode. To induce stress and calm mental states, a Stroop Color Word Task (SCWT) was utilized. EEG data were analyzed by extracting the mean power of 4 frequency bands using the Fast Fourier Transform (FFT). The different machine learning classifiers: K-Nearest Neighbors (KNN), Discriminant Analysis (DA), Naive Bayes (NB), Decision Tree (DT), and Support Vector Machine (SVM), were then compared for accuracy, sensitivity, and specificity. The behavioral result showed that the accuracy of detecting the SCWT under stress was reduced by 50%. The SVM outperformed other classifiers, achieving the highest classification performance in subject dependent with 99.98 ± 0.09, 99.96 ± 0.122, 99.85 ± 0.30, and 99.27 ± 0.57 accuracies in alpha, beta, theta, and delta band, respectively. In conclusion, alpha and beta bands showed a slightly higher accuracy than other frequency bands. Meanwhile, SVM outperformed other classifiers, achieving the highest classification accuracy of 99.98 % with the mean power of the alpha band.