Classification of Motor Tasks from EEG Signals Comparing Preprocessing Techniques

The electroencephalogram (EEG) has been used to control non-invasive brain-computer interface (BCI). EEG Signal is very susceptible to artifact that can interfere on the performance of the classifiers used in BCI system. There are many methods used to identify, reject, and remove artifacts. However, no consensual standard metrics for performance evaluation of these methods is available. The aim of this work is to study the performance of the different preprocessing techniques in classification, using raw EEG data and power spectra. Here, the preprocessing is the bandpass filtering, filtering and artifact removal by Independent Components Analysis (ICA), and filtering and rejection of artifacts by threshold. EEG signals from six right-handed healthy volunteers were divided in three tasks: observation (elbow flexion and extension); elbow flexion movement; elbow extension movement. According to the results without feature extraction (raw EEG), filtering and artifact removal by ICA had better accuracy. In addition, with feature extraction (power spectra), the bandpass filtering is preferred because of simplicity and no loss of data, even if it showed a slightly worse performance.