标题:M(k-NN)G-DPC: density peaks clustering based on improved mutual K-nearest-neighbor graph
作者:Fan, Jian-cong; Jia, Pei-ling; Ge, Linqiang
作者机构:[Fan, Jian-cong; Jia, Pei-ling] Shandong Univ Sci & Technol, Coll Comp Sci & Engn, Qingdao, Shandong, Peoples R China.; [Fan, Jian-cong] Shandong Un 更多
通讯作者:Fan, JC;Fan, JC
通讯作者地址:[Fan, JC]Shandong Univ Sci & Technol, Coll Comp Sci & Engn, Qingdao, Shandong, Peoples R China;[Fan, JC]Shandong Univ Sci & Technol, Prov Key Lab Info 更多
来源:INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
DOI:10.1007/s13042-019-01031-3
关键词:Clustering; Mutual k-nearest-neighbor graph; Density peak
摘要:Clustering by fast search and detection of density peaks (DPC, Density Peaks Clustering) is a relatively novel clustering algorithm published in the Science journal. As a density-based clustering algorithm, DPC produces better clustering results while using less parameters than other relevant algorithms. However, we found that the DPC algorithm does not perform well if clusters with different densities are very close. To address this problem, we propose a new DPC algorithm by incorporating an improved mutual k-nearest-neighbor graph (M(k-NN)G) into DPC. Our M(k-NN)G-DPC algorithm leverages the distance matrix of data samples to improve the M(k-NN)G, and then utilizes DPC to constrain and select cluster centers. The proposed M(k-NN)G-DPC algorithm ensures an instance to be allocated to the fittest cluster. Experimental results on synthetic and real world datasets show that our M(k-NN)G-DPC algorithm can effectively and efficiently improve clustering performance, even for clusters with arbitrary shapes.
收录类别:SCIE
资源类型:期刊论文
TOP