Two-Phase Spectral Clustering Based on Discretization

Abstract

As spectral clustering has become increasingly popular in recent years, further research on it is very important. Due to the important role of discretization in data mining, a new clustering approach integrating discretization with density-sensitive spectral clustering namely density-sensitive spectral clustering of categorized data (DSSCCAT) is proposed. To alleviate the high computational complexity of DSSCCAT, two-phase spectral clustering (TPSC) algorithm is proposed, which involves two phases: construct the representatives of the original dataset and cluster the representatives with DSSCCAT. Experimental results on UCI datasets show the feasibility of combining discretization with density-sensitive spectral clustering. TPSC can obtain desirable clusters with high performance. Furthermore, TPSC outperforms DSSCCAT obviously in terms of computational time.