Density Peaks Clustering Based on Candidate Center and Multi Assignment Policies

Density peak clustering (DPC) the center selection ignores environment of samples, which can easily result in incorrect identification cluster centers. Additionally, a single allocation approach produce joint effect mistakes. To address above issues, density peaks based on candidate and multi assignment policies (DPC-CM) is proposed. Firstly, to retain centers with relatively small densities, DPC-CM introduces thresholds delineate regions decision graph. It also incorporates shared nearest neighbor definitions ensure proper selection. Then, results backbone are corrected using information, suppressing cascading errors effectively. Finally, points used as basis for label propagation improves efficiency while taking into account surroundings sample sites situated. The fault tolerance algorithm has been enhanced by multi-step strategy. Experimental synthetic UCI datasets validate that outperforms comparative algorithms complex morphological unevenly distributed datasets.