A Top-Down Partitional Method for Mutual Subspace Clusters Using K-Medoids Clustering

摘要

In most of the applications, data in multiple data sources describes the same set of objects. The analysis of the data has to be carried with respect to all the data sources. To form clusters in subspaces of the data sources the data mining task has to find interesting groups of objects jointly supported by the multiple data sources. This paper addresses the problem of mining mutual subspace clusters in multiple sources. The authors propose a partitional model using k-medoids algorithm to determine k-exclusive subspace clusters and signature subspaces corresponding to multiple data sources, where k is the number of subspace clusters to be specified by the user. The proposed algorithm generates mutual subspace clusters in multiple data sources in less time without the loss of cluster quality when compared to the existing algorithm.