Feature selection through binary brain storm optimization

Abstract Feature selection stands for the process of finding the most relevant subset of features based on some criterion, which turns out to be an optimization task. In this context, several metaheuristic techniques have been extensively studied achieving results comparable to some state-of-the-art and traditional optimization techniques. This paper introduces a variation of the Brain Storm Optimization (i.e., Binary Brain Storm Optimization) for feature selection purposes, where real-valued solutions are mapped onto a boolean hypercube using different transfer functions. The proposed Binary Brain Storm Optimization was evaluated under different scenarios and with its results compared to some state-of-the-art techniques. Its overall performance presented suitable results that are comparable to the other techniques, thus showing to be a promising tool to the problem of feature selection.