Self-adaptive two-stage density clustering method with fuzzy connectivity

Abstract

Density Peak Clustering (DPC) was proposed in the journal Science in 2014 and has been widely applied in many fields due to its simplicity and effectiveness. However, there are few studies on the effectiveness of DPC algorithm and its variants on non-clean data sets. Inspired by the idea that DPC algorithm combines density and distance when determining clustering center, this paper creatively designs a two-stage density clustering method with fuzzy connectivity (TS-DCM). It could be used to distinguish different cluster partitions and further identify noise points and sample points. In addition, this paper also introduces a new clustering index: fuzzy connectivity, which could not only adjust the selection of DPC cutoff distance, but also provide a reference for adaptive adjustment of TS-DCM parameter selection, greatly improving the operating efficiency of the clustering algorithm. At the same time, a self-adaptive two-stage density clustering method (STS-DCM) is proposed to adjust the selection of parameters according to the feedback of clustering results. Finally, compared with other traditional and popular clustering algorithms, it is verified that the proposed algorithm has significant advantages in speed and accuracy. Moreover, for non-clean data sets, the algorithm is robust and effective.