Machine Learning-Based Stress Level Detection from EEG Signals

Recent statistical studies indicate an increase in mental stress in human beings around the world. Due to the recent pandemic and the subsequent lockdowns, people are suffering from different types of stress for being jobless, financially damaged, loss of business, deterioration of personal/family relationships, etc. Stress could be a severe factor for many common disorders if experienced for a long time. Stress is associated with the brain activities of human beings that can be scanned by electroencephalogram (EEG) signals which is very complex and often challenging to understand the signal’s pattern. This paper presented a system to detect the stress level from the EEG signals using machine learning algorithms. The proposed method, at first, removed physiological noises from the EEG signal applying a band-pass FIR filter. A discrete wavelet transform (DWT) method was used for features extraction from the filtered EEG signal. The features were classified using a set of classifiers those are k-nearest neighbors (kNN), support vector machine (SVM), Naïve Bayes, and linear discriminant analysis (LDA). Two levels of stressed EEG data were considered and found the classification accuracy of 86.3%, 91.0%, 81.7%, and 90.0%. The highest classification accuracy, the SVM classifier, outperforms the current state of the art by 15.8%.