Classification and Quantitative Estimation of Cognitive Stress from In-Game Keystroke Analysis using EEG and GSR

This paper presents a novel approach to classify cognitive stress, at an inter-personal level, induced by a proposed Stroop-Tetris game using Electroencephalogram (EEG), Galvanic Skin Response (GSR) and Photoplethysmogram (PPG). Features are derived from each sensor aimed at discriminating low and high cognitive stress, followed by feature reduction. Leave-one-subject-out cross-validation across the data shows high median accuracies in classification individually for EEG, GSR and PPG analysis (72.75%, 69.30% and 64.75% respectively), which was further enhanced to 77.25% by employing feature level fusion. A novel metric derived from the keystrokes of the subjects is introduced to quantify realtime stress. An intra-personal regression model to relate EEG features with the real-time stress metric demonstrates high correlation and low p-value between predicted and actual values of this metric.