Three — Channel electro-encephalogram (EEG) signal analysis by independent component analysis and classification by linear discriminant analysis

Electro-encephalogram (EEG) based brain-computer interface (BCI) has aim to associate neurological phenomenon (It is the source of control for any BCI system) with target action (Intentional state). The association with target is highly dependent on signal processing methodology as well as selected features from brain signals. This paper proposes a novel pre-processing method for EEG based BCI system, which includes second order statistics (SOS) based on single time lagged covariance matrix of brain signals. AMUSE algorithm is a part of blind source separation (BSS) algorithm, which includes SOS based on single time lagged covariance matrix. To evaluate the effectiveness of this preprocessing method (AMUSE Filtering) linear discriminant classifier (LDC) is adopted to classify the Graze BCI data set which was used in BCI competitions 2003. The selected feature for classification is power spectral density (PSD) in frequency band 8-30Hz. The performance of a proposed method has been evaluated in the terms of classical evaluation criteria classification accuracy (ACC), error rate (ERR) and modern evaluation criteria mutual Information (MI) and Cohen´s Kappa coefficient (κ).