MRMR based feature selection for the classification of stress using EEG

Abstract

Mental stress is a social concern causing functional disability during work routines. The evaluation of stress using electroencephalogram signals is a topic of contemporary research. EEG provides several different features and the selection of appropriate features becomes a question. This study presents the utilization of feature selection using maximum relevance and minimum redundancy (MRMR) based on mutual information (MI) on the obtained features from electroencephalogram (EEG) signals during stress and control tasks. We moved forward in recording EEG during stress which was induced by taking up an eminent experimental model based on the Montreal Imaging Stress Task (MIST). The induced stress was endorsed by the performance during the task and the response of the subjects. The methodology consist of EEG feature extraction such as the absolute power and relative power, feature selection (MI) and classification using the support vector machine. The results of the proposed methodology showed a maximum accuracy of 93.75% and above 85% accuracy throughout the experiment. The performance is better than the existing studies in the literature. In conclusion, the MRMR criterion of feature selection using MI gives reliable and consistent results for the classification of stress.