Intelligent Hybrid Swarm based Feature Selection Methods using Rough Set

Abstract

New feature selection methods based on Rough Set and hybrid optimization technique are proposed in this paper. In this work Feature Selection (Feature Reduction) has been implemented using Rough Set. Lower approximation based Rough Set has been used to calculate Positive Region which is consequently used to calculate Rough Dependency measure. Weighted sum of rough dependency measure and difference of total features of dataset and reduct normalized with respect to total feature, is used as fitness function. To optimize (maximize) this fitness function, a hybrid method of swarm intelligence algorithms like Intelligent Dynamic Swarm (IDS) and Particle Swarm Optimization (PSO) has been proposed and new method of population initialization has also been proposed. This method has been implemented on UCI repository based benchmark datasets of and it is shown that it results in improved reducts in terms of number of features, execution time with acceptable classification accuracy.