An efficient density-based clustering algorithm with circle-filtering strategy

Abstract

Recently a density peaks clustering algorithm (DPC) was proposed to obtain arbitrary shapes of the clusters effectively. The cluster centres are discovered by finding density peaks according to the decision graph which drawn based on the density-distance. However, the computational complexity is extremely high for calculating the density-distance of each point, which limits the application of DPC for the large-scale data sets. To overcome this limitation, an efficient density-based clustering algorithm with circle-filtering strategy (CFC) is proposed. CFC removes useless points based on a circle-filtering strategy first, and then the cluster centres are selected according to the remaining points. Experimental results show that CFC can reduce the computational complexity on the basis of ensuring the accuracy of clustering, and outperforms DPC.