Emotion Identification Based on EEG Rhythms Separated using Improved Eigenvalue Decomposition of Hankel Matrix

This paper presents a framework for identifying emotional state of humans using their electroencephalogram (EEG) signals. The accurate and efficient identification of multiple classes of emotion using non-stationary EEG signals is a challenging task. The emotion identification framework can be developed by applying proper signal processing and machine learning algorithms on the EEG signals. The improved eigenvalue decomposition of Hankel matrix (IEVDHM) is used to decompose the EEG signal into various components. The rhythms are obtained by adding the components together whose mean frequency falls in the range of the respective rhythm. Then from each rhythm, two features are computed namely, permutation min-entropy and Katz fractal dimension measures. Using ensemble subspace k-nearest neighbor (KNN), the feature values are classified into different emotional states namely, happy, sad, fear, and neutral. For testing the performance of the proposed framework, the 10-channel EEG signals were recorded from 39 subjects which include 19 females and 20 males. The average accuracy of 91.30 % is obtained for the proposed framework for human emotion identification using EEG signals. Such emotion identification algorithm can help in obtaining emotional state of the user in brain-computer interface (BCI).