An elitist binary PSO algorithm for selecting features in high dimensional data

Abstract

An elitist model of Binary Particle Swarm Optimization (BP SO) algorithm is proposed for feature selection from high dimensional data. Since the data are highly redundant, a fast pre-processing algorithm is employed to reduce features from high dimensions in a crude manner. The reduced feature subsets being still high dimensional, a further reduction is achieved by the proposed algorithm. The non-dominated sorting PSO algorithm is performed on the combined solutions of each two successive generations that also help to preserve the best solutions in a generation. The fitness functions are suitably formulated in multi objective framework for the conflicting objectives, i.e., to reduce the cardinality of the feature subsets and to increase the accuracy. The performance of the proposed algorithm is demonstrated on three high dimensional benchmark datasets, i.e., colon cancer, lymphoma and leukemia data. © Springer International Publishing Switzerland 2014.