Title: Unsupervised bidirectional feature selection based on contribution entropy for medical databases

Authors: D. Devakumari, K. Thangavel, K. Sarojini

Addresses: Department of Computer Science, Government Arts College, Dharmapuri 636705, Tamil Nadu, India. ' Department of Computer Science, Periyar University, Salem 636011, Tamil Nadu, India. ' Department of Computer Applications, S.N.R. Sons College, Coimbatore 641006, Tamil Nadu, India

Abstract: Feature selection is one of the important pre-processing steps in data mining for selecting informative feature subsets in large noisy data sets. This paper proposes an unsupervised feature selection method known as bidirectional selection based on the contribution entropy of individual features. The proposed feature selection method was tested on benchmark medical data sets, and the quality of the clusters obtained was evaluated using the homogeneity and separation ratio. Results show an improvement in cluster quality when compared with existing feature selection methods.

Keywords: singular value decomposition; SVD; contribution entropy; forward selection; backward elimination; bidirectional feature selection; clustering; homogeneity; separation; medical databases; data mining; cluster quality.

DOI: 10.1504/IJHTM.2011.042368

International Journal of Healthcare Technology and Management, 2011 Vol.12 No.5/6, pp.364 - 378

Published online: 28 Mar 2015 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article