A Comparative Analysis of Different Feature Extraction Techniques for Motor Imagery Based BCI System

Abstract

Brain Computer Interface (BCI) system based on Motor Imagery (MI) signals is one among the most prospering systems in the field of BCI. Motor imagination does not involve motor output from the human. This can be used to help motor disabled people to accomplish elementary tasks by themselves. One of the primary difficulties facing MI-based BCIs is extracting discriminative features from the EEG signal. This study seeks to minimize the number of features input to classifier by finding out the best feature extraction techniques that can achieve high system accuracy with minimum computation cost. To achieve this purpose, this study compares five different feature extraction techniques; Root Mean Square (RMS), Renyi entropy, Shannon entropy, Katz fractal dimension and Common Spatial Patterns (CSP), regarding accuracy and execution time. To ensure that these techniques are reliable; they are applied on three benchmark datasets; BCI Competition III datasets IVa and IIIa, and BCI Competition IV dataset IIa. The extracted features are examined with two classifiers; Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA). Experimental results showed that CSP is the best feature extraction technique compared to other examined techniques, yet Renyi entropy has the least computation time which is a critical issue for an online system.