Hierarchical Subspace Clustering

摘要: It is well-known that traditional clustering methods considering all dimensions of the feature space usually fail in terms efficiency and effectivity when applied to high-dimensional data. This poor behavior based on fact clusters may not be found space, although exist subspaces space. To overcome these limitations methods, several for subspace have been proposed recently. Subspace algorithms aim at automatically identifying lower dimensional which exist. There two types algorithms: Algorithms detecting axis-parallel and, as an extension, finding are arbitrarily oriented. Generally, hierarchically nested, i.e., low dimensionality form a cluster higher dimensionality. Since existing able detect complex structures, hierarchical approaches applied. The goal this dissertation develop new efficient effective by novel challenges approach proposing innovative solid solutions challenges. The first Part work deals with analysis subspaces. Two search simultaneously arbitrary order hierarchies clusters. Furthermore, visualization model result means graph representation provided. In second oriented discussed. The so-called correlation can seen extension clustering. Correlation aims grouping data set into subsets, clusters, such objects same show uniform attribute correlations. combine density-based Principal Component Analysis identify clusters. The last addresses interpretation results obtained from algorithms. A general method introduced extract quantitative information linear dependencies between given models used predict probability object created one models. Both, effectiveness presented techniques thoroughly analyzed. benefits over shown evaluating synthetic well real-world test sets.