Meta heuristic assisted automated channel selection model for motor imagery brain computer interface

Abstract

Recently, the practical Motor Imaginary Brain-computer interface (MIBCI) are being largely developed. In general, these systems adopt high-density ElectroEncephaloGraphy (EEG) channels with lack of channel optimization. The data processing complexity increases due to the artifacts and noise in the several channels and this tends to lessen the classification performance. Hence to address this problem, a novel channel selection approach will be developed. In this work, the most appropriate channel will be selected by a new hybrid optimization approach Geometric Mean based Moth flame hybridized with Fire fly Update (GM-MFU) model. This is the hybridized form of standard “Moth-Flame Optimization (MFO) algorithm and Firefly (FF) optimization algorithm”. In addition to this, the appropriate samples corresponding to the channels are also selected with the proposed GM-MFU model. An objective model is developed for congruent selection of the optimal channels and solution samplesThe selected channels as well as features extracted from the original signal with “Common Spatial Pattern(CSP)” are subjected for classification via Bayesian Linear Discriminate Analysis (BLDA). The final resultant from Bayesian Linear Discriminate Analysis (BLDA) provides the classified Motor Imagery (MI) EEG signal. The proposed channels election approach with hybridized meta-heuristic optimization is evaluated over the state of the art models in terms of positive, negative measures and Kappa coefficient as well.