Single-Trial EEG Classification of Movement Related Potential

Abstract

A single trial electroencephalogram (EEG) classification system is proposed for left/right self-paced tapping discrimination. Features are extracted from theta, mu and beta rhythms and readiness potential (Bereitschaftspotential) that precede the voluntary movement. Feature extraction relies on regression fitting and wavelet decomposition. These two approaches are compared through two linear classification functions, a Fisher linear discriminant and a minimum-squared-error linear discriminant function. We show that discrete wavelet decomposition is an effective tool for both EEG frequency component separation and feature extraction, and therefore suitable for pre-movement left/right discrimination. The algorithms are applied to the data set of the "BCI Competition 2001" with a classification accuracy of 96%.