Feature-Weighted Fuzzy K-Modes Clustering

Abstract

Fuzzy k-modes (FKM) are variants of fuzzy c-means used for categorical data. The FKM algorithms generally treat feature components with equal importance. However, in clustering process, different feature weights need to be assigned for feature components because some irrelevant features may degrade the performance of the FKM algorithms. In this paper, we propose a novel algorithm, called feature-weighted fuzzy k-modes (FW-FKM), to improve FKM with a feature-weight entropy term such that it can automatically compute different feature weights for categorical data. Some numerical and real data sets are used to compare FW-FKM with some existing methods in the literature. Experimental results and comparisons actually demonstrate these good aspects of the proposed FW-FKM with its effectiveness and usefulness in practice.