Fast main density peak clustering within relevant regions via a robust decision graph

Junyi Guan, Sheng Li, Jinhui Zhu, Xiongxiong He, Jiajia Chen

Abstract: Although Density Peak Clustering (DPC) can easily locate cluster centers by detecting density peaks in its decision graph, its allocation strategy may unadvisedly associate irrelevant points, its decision graph may mislead the cluster center selection, and its high computational complexity 𝑂(𝑛2) shies itself away from large- scale data. Herein, a Fast Main Density Peak Clustering Within Relevant Regions Via A Robust Decision Graph (R-MDPC) is proposed. R-MDPC assigns points within the relevant regions to avoid the association of irrelevant points. With the removal of regional differences and the attenuation of satellite peaks, a robust decision graph is obtained. Moreover, based on the kNN distance of data points, R-MDPC is believed to be suitable for large-scale data. Experimental results demonstrated the high robustness of R-MDPC’s decision graph in identifying cluster centers, and its outstanding performance and fast running speed in recognizing complex-shaped clusters.