EEG-based Mental Workload Detection in a Mental Arithmetic Task using Machine Learning

Abstract

The increasing workload in day-to-day activities has aggravated an already poor mental health of this generation. There is an urgent requirement to address this issue which often tends to be overlooked. In this paper, we propose an efficient methodology to detect mental workload during a mental arithmetic task from EEG signals. We utilise the publicly available ‘EEG During Mental Arithmetic Tasks’ dataset to discriminate between rest and active mental phases. We extract a feature pool containing Spectral Density, Relative Power, Alpha/Beta ratio, and Theta/Alpha ratio features belonging to the frequency domain. We use the Recursive Feature Elimination with Cross-Validation (RFECV) feature selection algorithm to select the significant set of features from the feature pool. Finally, we test out the efficiency of the reduced set of features using three classifiers- Naive Bayes, Support Vector Machine (SVM) and K-Nearest Neighbors (K-NN). We obtain our best accuracy of 97.5% using the SVM classifier, outperforming several state-of-the-art methodologies on the particular dataset.