Clustering by Finding Average Density

Abstract

Density Peak Clustering (DPC) algorithm can get better clustering results of data sets with lower dimensions. However, for the high-dimensional data sets, there are many nodes in the clustering center area; their densities are relatively high and close to each other, so that it is hard to identify the centering node accurately. It is found that the accuracy of the traditional DPC algorithm decreases terribly with the increasing of data set dimensions. In order to deal with the indistinguishable density problem, we propose a new clustering method, called Clustering by Finding Average Density (CFAD), which can enhance the clustering effect of high-dimensional data sets. Experiments show that the proposed algorithm outperforms DPC for both the artificial and the real data sets.