Research on Density Sensitive Clustering Algorithm for Non-convex Sets

Abstract. Applying Clustering to non-convex data is a challenging task, and traditional clustering algorithms often fail to achieve good results. In this paper, an improved spectral clustering algorithm based on density sensitivity (DSISC algorithm) is proposed. By using the ensemble selection strategy for the mean shift algorithm, relatively good optional clusters are selected from the nonconvex data sets, and then the number of clusters is transported into the spectral clustering algorithm as input, and the density-sensitive distance is used as the similarity measure. The experimental results give us clear information that the DSISC is better than traditional mean shift algorithm and spectral clustering algorithms in normalized mutual information clustering error rate.