Feature Selection Using EEG Signals: A Novel Hybrid Binary Particle Swarm Optimization

Abstract

As the tendency to use robots in everyday life is constantly growing, Human-Robot Interaction (HRI) is a significant and promising field of research. The direct link between humans and robots is studied by brain robot interaction area and the most popular non-invasive mean to record brain activity is through EEG signals. In this paper, we propose a novel hybrid Binary Particle Swarm Optimization (BPSO) algorithm which embeds k-means clustering method to enhance feature selection accuracy and computational cost. This effort could be implemented in a variety of HRI applications such as controlling a smart wheelchair with brain signals. In order to address the problem of trapping in local minimum, a novel adaptive mutation rule was introduced in the scheme of the BPSO algorithm. To evaluate the performance of the proposed scheme, an EEG motor imagery dataset from GigaScience database including 50 subjects was used. Pre-processing and feature extraction were performed using various methods to yield an extensive set of features. Finally, the proposed algorithm showed 5.7% and 4.6% mean accuracy enhancement in S-shaped and genotype-phenotype BPSO algorithm to achieve 88.5% and 91.5% mean accuracy, respectively.