Assessment of Mental Workload in Real-Life Setup using EEG Synchronization Measures

Abstract

EEG data from prefrontal channels AF7 and AF8 acquired using the consumer-oriented MUSE-S device were analyzed for the classification of mental workload levels during an n-back memory game. 30 subjects were enrolled to the study. Each recording session contained 9 games of 3 different levels. The feature set included features based on magnitude-squared coherence, spectral entropy, phase locking value and the newly introduced coherence entropy. The AdaBoost classifier was used to evaluate the performance of the feature sets. In 3-class classification, accuracy of 0.93 was obtained with the full feature set. It was found that game levels 0 and 2 were more difficult to discriminate compared to levels 0 and 1. Besides the full feature set, spectral entropy based features also showed outstanding performance.