Mining Significant Subspaces
Anthony J.T. Lee, Ming‐Chih Lin, Yun-Ru Wang, Kuo-Tay Chen · 2010
As both the number of dimensions increases, existing clustering methods in full feature space are not appropriate to cluster data in databases. Thus, the subspace clustering has attracted more and more attention recently. In this paper, we propose a novel method to mine significant subspaces from all frequent subspaces, where a subspace is frequent if it contains enough data points. The proposed method consists of three phases. First, we generate all frequent 2-dimensional subspaces. Second, we recursively combine frequent k-dimensional subspaces to generate frequent (k+1)-dimensional subspaces, k≥2. Finally, we adopt a greedy algorithm to summarize the frequent subspaces generated and select the significant ones. The experimental results show that the proposed method has better quality and coverage than DUSC, and better quality than FIRES.