Optimizing channel selection using multi-objective FODPSO for BCI applications

Abstract

ABSTRACT Brain-computer interfacing (BCI) requires multichannel electroencephalogram (EEG) or electrocorticogram (ECoG) signal acquisition for better performance. The selection of optimum channels that provides the best accuracy is itself a problem to solve because noisy/irrelevant channels can complexify the system and degrade its performance. This paper presents a novel automated model for optimum channel selection for BCI applications using the Fractional Order Darwinian Particle Swarm Optimization (FODPSO). To assess the information content of selected channels, a Support Vector Machine (SVM) classifier is used. The weighted sum of the number of channels and the classification accuracy on validation samples is taken as the fitness value for the FODPSO based binary optimization method (where all the variables are binary). The FODPSO and SVM based algorithm is evaluated on the electrocorticography (ECoG) recordings that have been used in BCI competition III. Dual tree complex wavelet transform (DTCWT) is used for preprocessing of the data and sample entropy is calculated for obtaining the most informative features from the preprocessed data. Eight channels are selected using this algorithm, yielding a classification accuracy of 0.81 for the testing dataset (classifying the ECoG signals of imagined movement of little finger and tongue) that compares favorably with the already reported methods. Experimental results successfully demonstrate that this channel selection algorithm works better both in terms of classification accuracy and in the reduction of the number of required channels.