Detection and Visualization of Subspace Cluster Hierarchies

摘要: Subspace clustering (also called projected clustering) addresses the problem that different sets of attributes may be relevant for clusters in high dimensional feature spaces. In this paper, we propose algorithm DiSH (Detecting cluster Hierarchies) improves following points over existing approaches: First, can detect subspaces significantly dimensionality. Second, uncovers complex hierarchies nested subspace clusters, i.e. lower-dimensional are embedded within higher-dimensional clusters. These do not only consist single inclusions, but also exhibit multiple inclusions and thus, modeled using graphs rather than trees. Third, is able to size, shape, density. Furthermore, visualize by means an appropriate visualization model, so-called graph, such relationships between explored at a glance. Several comparative experiments show performance effectivity DiSH.