Multiplex neighbor density peaks clustering for uneven density data sets

The local density of density peaks clustering (DPC) algorithm ignores the density difference of the data with uneven density distribution, which easily leads to the cluster centers found in the dense area resulting in poor clustering effect. In order to overcome the above shortcomings, this paper proposes multiplex neighbor density peaks clustering for uneven density data sets (MN-DPC). Firstly, the natural nearest neighbor information is used to define the local density of samples to balance the density difference between samples in sparse and dense regions, so as to correctly find the class cluster centers in sparse regions; Secondly, the sample similarity is weighted by using the shared nearest neighbor and natural nearest neighbor information, which strengthens the similarity between samples of the same type of cluster and effectively avoids the misallocation of samples in sparse regions. This paper compares the MN-DPC algorithm with the IDPC-FA, DPC-DBFN, DPCSA, FNDPC, FKNN-DPC and DPC algorithms. The experimental results show that the MN-DPC algorithm can effectively cluster data sets with uneven density distribution and UCI data sets. © 2024 South China University of Technology. All rights reserved.