期刊名称：International Journal of Computer Science and Information Technologies
电子版ISSN：0975-9646
出版年度：2014
卷号：5
期号：3
页码：3839-3842
出版社：TechScience Publications
摘要：Clustering high dimensional data is an emerging research field. Most clustering technique use distance measures to build clusters. In high dimensional spaces, traditional clustering algorithms suffers from a problem called “curse of dimensionality”. Subspace clustering groups similar objects embedded in subspace of full space. Recent approaches attempt to find clusters embedded in subspace of high dimensional data. Most of the previous subspace clustering works discovers subspace clusters, by regarding the clusters as regions of higher densities. The regions are identified dense if its density exceeds the density threshold. As the cluster densities varies in different subspace cardinalities, it suffers from a problem called “density divergence problem”. We follow the basic assumptions of previous work DENCOS. It is found that varying region densities are used to overcome density divergence problem. All previous approaches are based on bottom-up method. In this paper a novel data structure is used which works on both bottom-up & top-down fashion. Performance results of this new novel data structure shows very good results and the efficiency outperforms the previous works.
关键词：Subspace Clustering; High dimensional data;Mining frequent patterns; Top down; Bottom Up
