Efficacy of Tsallis Entropy in Clustering Categorical Data

Abstract

Categorical data clustering is an important area of research today as databases usually contain categorical data [1]. The current work proposes that the behavior of attributes in categorical dataset is important in selecting the clustering algorithm. A Tsallis entropy based categorical data clustering (TEC) algorithm is also presented. It is shown that when the attributes depict power law behavior, the proposed TEC algorithm outperforms existing Shannon entropy based clustering algorithms. Experimental results on UCI and WEB KB datasets validates the efficacy of TEC algorithm.