The Influence of Smoothing Filtering Methods on the Performance of an EEG-Based Brain–Computer Interface

Electroencephalography (EEG) recording is highly vulnerable to physiological or technical artifacts, which may reduce the performance of EEG-based brain–computer interface (BCI) systems. A number of noise artifact removal methods have been used to overcome this issue and hence estimate the cognitive state information much better, which will lead to the design of EEG-based BCI systems that are more practical, reliable, and accurate. Smoothing filter techniques are popularly used to remove noise and retain the morphology of signals. The purpose of this study is to compare three smoothing filters—median, Savitzky-Golay, and regularization—in the analysis of EEG data. To do so, we used publicly available motor imagery and P300 datasets to evaluate the effects of applying the aforementioned smoothing filters on the classification of right- versus left-hand imagery movements and target versus nontarget characters in spellers, respectively. The results show that smoothing EEG by regularization increased the coefficient of determination ( $r^{2}$ ) values between the target and nontarget responses and slightly improved the signal-to-noise ratio relative to the other smoothing filters. Moreover, the results show that power spectral density for EEG smoothed by regularization reveals more discriminative information about left- and right-hand imagery movements. The classification results show that smoothing EEG by regularization provides the best classification accuracy in both datasets.