Applications of adaptive and non-adaptive filters for classification of eeg signals of motor imagery

Brain Computer Interface(BCI) is based on the classification of human intentions or intentional thought without the need for any physical ability. In BCI studies, signals which obtained using Electroencephalography(EEG) exposed to some environmental or internal noise (eye movements, ECG, etc.). One of the artifacts that can significantly affect classification accuracy is electrooculogram(EOG). In this study, BCI competition IV dataset 2a was used which was organized by the University of Graz. The data set with 4 classes was reduced to 2 classes and only the right and left motor imagery data were classified according to the offline paradigm. Simultaneously with EEG data from 22 channels, EOG data were obtained from 3 electrodes mounted around the eye. Conventional bandpass filters and recursive least square (RLS) adaptive filters are used to remove EOG noise from EEG signals. Feature vectors extracted from filtered signals by CSP method were applied to the inputs of linear discriminant analysis (LDA), support vector machines (SVM), naive bayes (NB) and k-NN classification algorithms. The combination of Chebyshev type 2 filter and SVM achieved the highest classification performance with an average accuracy of 72%. Furthermore, the classification results obtained with the RLS adaptive filter were compared with those of the bandpass filters. The combination of the RLS algorithm and the LDA classifier showed lower performance than traditional bandpass filters with a 64% accuracy rate. The findings of the study may shed light on future studies. Keywords: Adaptif Filter, Recursive Least Square, Brain Computer Interface, Electrooculogram, Electroencephalography, Traditional Bandpass Filters, Common Spatial Patterns