Automatic Clustering simultaneous Feature Subset Selection using Differential Evolution

Abstract

Clustering is important and widely used in variety of machine learning applications. High dimensionality is a curse to clustering, that declines the algorithm performance in knowledge discovery and increases the algorithm complexity. The high dimensionality risk can be reduced with proper selection of good subset of features by avoiding irrelevant features. Selection of good clusters with proper subset of features is posed as a problem of optimization, can be solved with powerful Meta heuristic methods. Till date, a number of evolutionary based solutions are available for both problems automatic clustering and feature selection. From a decade, Automatic Clustering using Differential evolution is found to be one of the successful methods in automatic clustering considering all features. There is no algorithm that finds optimal clusters with simultaneous feature sub set selection. The paper proposes a novel Automatic Clustering with simultaneous Feature Subset Selection using Differential Evolution (ACFSDE) algorithm. ACFSDE is an enhanced variant to ACDE, defines a new chromosome structure for selection of optimal features and/for optimal clusters. Experiments are conducted in two fold; one is, using numeric UCI benchmark datasets and synthetic data sets. Second is to study the performance of ACFSDE for texture image segmentation by applying on images. The results on numeric data are evaluated using six clustering validity measures and are compared with five other existing clustering algorithms. The ACFSDE results are very prominent with more than 80% of average accuracy.