On the Effectiveness of ICA based Eye Artifact Removal from EEG Windows of Different Lengths

Eye artifacts, i.e., blinks and saccades, are usually nonavoidable when recording electroencephalogram (EEG) data. These artifacts can affect the performance of classifying the EEG patterns especially in real world applications, e.g. brain computer interfaces. To evaluate the effectiveness of independent component analysis (ICA) based eye artifact removal methods, the data are analyzed in batch and window-based modes in this paper. Despite the improvements achieved in the batch mode, it turns out that applying the removal methods to overlapping windows of the EEG data stream does not improve the classification performance.