Study of Different Filter Bank Approaches in Motor-Imagery EEG Signal Classification

Motor-imagery EEG signal classification is an important topic in the Brain-Computer Interface domain. In this chapter, two distinct variants of filter bank models have been used once in normal mode (filter is applied on the entire input signal) and again with overlapping and non-overlapping temporal sliding windows (filter is applied on the segments of input signal): (a) Filter bank of 4 Hz frequency band with cut-off frequencies 4–24 Hz, (b) Filter bank of five frequency bands- delta, theta, alpha, beta and gamma brain rhythms. Subsequently, the Common Spatial Pattern (CSP) has been implemented both on the each filtered output EEG signal as well as on the combined filtered output to form the final feature-sets for all the filter bank techniques. The traditional bagging ensemble classifier has been improvised using the Differential Evolution (DE)-based error minimization for the model training. The obtained classification accuracies are then compared with one another to examine the performance of the proposed approach. The best classification accuracy obtained from our study is 86.43%.