Distributed data clustering: A comparative analysis

Abstract

Due to explosion in the number of autonomous data sources, there is a growing need for effective approaches to distributed clustering. This paper compares the performance of two distributed clustering algorithms namely, Improved Distributed Combining Algorithm and Distributed K-Means algorithm against traditional Centralized Clustering Algorithm. Both algorithms use cluster centroid to form a cluster ensemble, which is required to perform global clustering. The centroid based partitioned clustering algorithms K-Means, Fuzzy K-Means and Rough K-Means are used with each distributed clustering algorithm, in order to analyze the performance of both hard and soft clusters in distributed environment. The experiments are carried out for an artificial dataset and four benchmark datasets of UCI machine learning data repository. © 2009 Springer-Verlag Berlin Heidelberg.