A novel subspace clustering algorithm with dimensional density

When clustering data of high dimension, most of the existing algorithms cannot reach people´s expectation due to the curse of dimensionality. In high-dimensional space, clusters are often hidden in subspaces of the attributes. The distribution of clusters is dense in the subspace and each attribute of the subspace. So objects belonged to the same subspace have similar density on each attribute. Based on this idea a novel subspace clustering algorithm SC2D is proposed. By introducing the definition of dimensional density, SC2D puts objects into the same cluster if they have similar dimensional density. And then clusters are separated from each other if there are more than one cluster in the same subspace. Experiments on both artificial and real-world data have demonstrated that SC2D algorithm can achieve desired results.