Feature Subset Selection for Clustering Using Binary Particle Swarm Optimization

Feature selection is one of the most important pre-processing steps when we consider a data mining, pattern recognition or machine learning problem. Finding an optimal feature subset, among all the combinations, is a NP-Complete problem. Despite lot of research in this domain, feature selection for clustering is still an unsolved issue. In this paper, a binary particle swarm optimization (PSO) algorithm has been proposed for feature selection in clustering. We aim at (i) Maximizing the Laplacian Score and (ii) Minimizing the inter-attribute correlation, and unifying the value using no preference method. Empirical studies have been conducted over 21 publicly available datasets. The average reduction is feature set cardinality is more than 71%. In terms of cluster validity also there is an efficiency of above 90%.