Hand Motion Identification Based on EEG Signals Classification

Abstract

Electroencephalogram (EEG) signals based Brain-Computer Interfacing is the latest trend for designing rehabilitation devices. This study critically compared the different classification approaches so that performance of EEG signals can be improved by using Support Vector Machine (SVM) classifier in conjunction with Independent Component Analysis (ICA), Principal Component Analysis (PCA), and Common Spatial Pattern (CSP) methods. Further, the performance of SVM was compared with the Minimum Distance Classifier (MDC) and Linear Discriminant Analysis (LDA) classifier. In this context, EEG signals were first acquired and filtered with a band-pass Butterworth filter whereas ICA was used for ocular artifact removal. Feature extraction was done by using the CSP method to enhance the classification accuracy which generates the discriminating feature variance and the PCA technique was employed for dimension reduction. The best classification accuracy (100%) was obtained from ten healthy subjects’ EEG dataset by using the SVM method followed by the LDA classifier. This study reveals that SVM classifier with ICA, CSP and PCA methods yielded the best result and able to enhance the practical implementation of rehabilitation robots.