Analysis of MLP and DSLVQ Classifiers for EEG Signals Based Movements Identification

Abstract

Brain-Computer Interfacing (BCI) is the latest research trend for developing the rehabilitation robotic system based on electroencephalogram (EEG) signals to make human life more comfortable. In this context, a framework was suggested to critically compare the performance of two different classification methods so that the performance of EEG signals could be improved in conjunction with Common Spatial Pattern (CSP), Independent Component Analysis (ICA) and Principal Component Analysis (PCA) approach. Further, the performance of Multilayer Perceptron Classifier (MLP) and Distinction Sensitive Learning Vector Quantization (DSLVQ) was compared with each other on a single feature accuracy scale. EEG dataset was recorded from ten healthy human subjects followed by band-pass Butterworth filtering for de-noising and ocular artifact rejection by ICA. The CSP was utilized for generating the discriminating features followed by PCA dimension reduction. After performing the all desired preprocessing steps, eight features were extracted to form the feature vector and classified by MLP and DSLVQ classifiers. The best classification accuracy of 98.75% was achieved with ten healthy subjects’ EEG datasets by exploiting the MLP method followed by the DSLVQ classifier. This study reveals that MLP classifier with PCA, CSP and ICA methods produced the best performance and able to enhance the practical implementation of various assistive robotic devices.