Comparison of Feature Extraction Methods for EEG BCI Classification

Abstract

This work analyzes several feature extraction methods used in today’s EEG BCI (electro-encephalogram brain computer interface) classification systems. Comparison of multiple EEG energy algorithms is presented for solving a 4-class motor imagery BCI classification problem. Furthermore, multiple feature vector generation techniques are employed into analysis. The effectiveness of CSP (common spatial pattern) filtering method in preprocessing step is shown. Channel difference feature extraction method is presented. It is discussed that key aim in today’s EEG signal analysis should be dedicated to finding more accurate techniques for determining better quality features. Initial tests prove that static feature extraction methods are not optimal and adaptive algorithms are required to overcome subject specific EEG signal variations. Further work and new dynamic feature extraction methods are required to solve the problem.