A Comparative Analysis of Different Preprocessing Pipelines for EEG-based Mental Stress Detection

Stress is considered one of the major reasons behind physical and mental health deterioration gradually. Long-term stress can also hamper the sleep cycle and people can suffer from severe depression. There are many existing methods to detect stress levels and Electroencephalogram (EEG) is one of them. It is a method for assessing stress levels by observing brain electrical activity. In this paper, mental stress was detected while the subjects were performing some mathematical calculations. We collected EEG signals from 23 subjects using 32 channel EEG headset. A critical part of this paper is to remove artifacts using both proposed and traditional pre-processing pipelines. Further, we extracted many frequency domain features and statistical features and incorporated several machine learning classification models. The accuracy found for the proposed pre-processing pipeline was 96.85% for Support Vector Machine (SVM) after 5-fold cross-validation whereas the accuracy for the traditional pre-processing pipeline was 87.05%.