Emotion Recognition via EEG Using Neural Network Classifier

TLDR: An efficient EEG signal analysis technique for human emotion classification using temporal and morphological features of EEG is presented and can be explored to develop an automated clinical application to assist patients suffering from stress disorders with more efficient classification rates.

Abstract: Automated assessment of human emotions via physiological signals has gained remarkable significance in the development of affective human–machine interfaces for stress detection. However, manual emotion analysis procedure is solely dependent upon the expertise of the analyst. An attempt has been made in this research to characterize and classify emotions from associated human neural responses via electroencephalography (EEG). Emotion-specific multichannel EEG dataset is acquired using 14-channel emotiv EEG neuroheadset. An efficient EEG signal analysis technique for human emotion classification using temporal and morphological features of EEG is presented. It uses power spectral and maximum/minimum peak features extracted from each EEG segment as an outcome to characterize emotion-specific EEG dynamics. A feed-forward neural network classifier is configured using Levenberg–Marquardt training algorithm to classify human emotions in two categories, viz, normal and stress states with classification accuracy of 60%. The experimental results reveal that the methodology adopted can further be explored to develop an automated clinical application to assist patients suffering from stress disorders with more efficient classification rates.