Kernel K-means for categorical data

Abstract

Clustering categorical data is an important and challenging data analysis task. In this paper, we explore the use of kernel K-means to cluster categorical data. We propose a new kernel function based on Hamming distance to embed categorical data in a constructed feature space where the clustering is conducted. We experimentally evaluated the quality of the solutions produced by kernel K-means on real datasets. Results indicated the feasibility of kernel K-means using our proposed kernel function to discover clusters embedded in categorical data. © Springer-Verlag Berlin Heidelberg 2005.