Leveraging EEG in Classifying Human Stress With Machine Learning Approach

Stress is one of mental disorder that may lead to negative consequences if not handled properly. Stress can trigger other mental disorders such as depression, anxiety disorders, even impulsive actions such as suicide if not treated immediately. Recognizing stress at an early stage can prevent the patient's condition from worsening, thereby increasing the patient's chances of recovery. The stress recognition process can be done by analyzing the fluctuation patterns of EEG data as brain wave signals which are able to reveal activity in the brain in various circumstances. Signal filtering with Bandpass Filtering and feature extraction with Fast Fourier Transform (FFT) will be carried out at the beginning of the process before training. The signals used are only signals that have relevant information about stress, verified through statistical analysis. Those data do not have a label especially if it is associated with stress. Thus, this study will propose a stress labeling mechanism through EEG signal strength analysis based on delta, theta, beta, and gamma signals. Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Decision Tree (DT) will be implemented as a classifier that is capable of classifying new unlabeled data. All models are trained using k-fold cross-validation. The results show that the highest accuracy model was SVM with a training and testing accuracy up to 1.00. SVM model is able to generalize very well on our EEG data in recognizing stress. Hence, this research succeeded in showing the empirical results that EEG data has a correlation with human stress. These findings can be used as a basis, especially for other researchers in similar domains in expanding understanding of the relationship between brain activity and stress experienced.