Features Selection Using Differential Evolution in Motor-Imagery Based Brain Machine Interface

Abstract

We exploit filter bank common spatial patterns (FBCSP) to extract raw features of EEG signals (using different frequency bands) and differential evolution (DE) algorithm to select optimal features. Since frequency bands vary from one subject to another and yield a large number of features, our mission is twofold: (i) overcome the curse of dimensionality; and (ii) select frequency bands that can lead to a better recognition performance. These two issues are addressed using differential evolution (DE) algorithm for feature selection. The results are compared to the six top results of the BCI competition IV. The proposed method is promising since it has outperformed the methods reported in the BCI competition IV 2b datasets.