Clustering by Searching Density Peaks via Local Standard Deviation

To solve the problem of DPC (Clustering by fast search and find of Density Peaks) that it cannot find the cluster centers coming from sparse clusters, a new clustering algorithms is proposed in this paper. The proposed clustering algorithm uses the local standard deviation of point i to define its local density \(\rho _i\), such that all the cluster centers no matter whether they come from dense clusters or sparse clusters will be found as the density peaks. We named the new clustering algorithm as SD_DPC. The power of SD_DPC was tested on several synthetic data sets. Three data sets comprise both dense and sparse clusters with various number of points. The other data set is a typical synthetic one which is often used to test the performance of a clustering algorithm. The performance of SD_DPC is compared with that of DPC, and that of our previous work KNN-DPC (K-nearest neighbors DPC) and FKNN-DPC (Fuzzy weighted K-nearest neighbors DPC). The experimental results demonstrate that the proposed SD_DPC is superior to DPC, KNN-DPC and FKNN-DPC in finding cluster centers and the clustering of a data set.