Nature-Inspired Algorithms for Selecting EEG Sources for Motor Imagery Based BCI

In this article we examine the performance of two well-known metaheuristic techniques (Genetic Algorithm and Simulating Annealing) for selecting the input features of a classifier in a BCI system. An important problem of the EEG-based BCI system consists in designing the EEG pattern classifier. The selection of the EEG channels used for building that learning predictor has impact in the classifier performance. We present results of both metaheuristic techniques on real data set when the classifier is a Bayesian predictor. We statistically compare that performances with a random selection of the EEG channels. According our empirical results our approach significantly increases the accuracy of the learning predictor.