Utilização de algoritmos de otimização por enxame aplicados a seleção de características

Abstract

The search for feature selection methods has been increasingly present in machine learning applications, especially in those where the number of available attributes is in the range of hundreds or even thousands. Such applications include, for example, word document processing, gene expression analysis, and combinatorial chemistry. Feature selection or selection of characteristics is a concept that proposes methods that aim to provide faster and more economical predictors, improve predictor prediction performance, and provide a better understanding of the underlying process that generated the data. Mathematically, feature selection is formulated as a combinatorial optimization problem. In general, addressing such problems in a way that finding the exact solution is not always feasible. In this way, computational intelligence methods can be used to allow the feature selection in practice. Therefore, the objective of this work is to present and propose optimization techniques guided by strategies of feature selection, among which we can highlight the optimization by swarm of particles, optimization of swarm by competition and the combination of both.