东北大学学报(自然科学版) ›› 2022, Vol. 43 ›› Issue (6): 761-768.DOI: 10.12068/j.issn.1005-3026.2022.06.001

• 信息与控制 •    下一篇

一种自适应的密度峰值聚类算法

马淑华, 尤海荣, 唐亮, 何平   

  1. (东北大学秦皇岛分校 控制工程学院, 河北 秦皇岛066004)
  • 修回日期:2021-05-21 接受日期:2021-05-21 发布日期:2022-07-01
  • 通讯作者: 马淑华
  • 作者简介:马淑华(1967-),女,河北秦皇岛人,东北大学秦皇岛分校教授.
  • 基金资助:
    国家自然科学基金资助项目(11705122); 河北省自然科学基金资助项目(F2020501040).

An Adaptive Density Peak Clustering Algorithm

MA Shu-hua, YOU Hai-rong, TANG Liang, HE Ping   

  1. School of Control Engineering, Northeastern University at Qinhuangdao, Qinhuangdao 066004, China.
  • Revised:2021-05-21 Accepted:2021-05-21 Published:2022-07-01
  • Contact: YOU Hai-rong
  • About author:-
  • Supported by:
    -

摘要: 针对密度峰值聚类(density peak clustering, DPC)算法不能根据数据集自适应选取聚类中心和截断距离dc,从而不能自适应聚类的问题,提出了一种自适应的密度峰值聚类(adaptive density peak clustering, ADPC)算法.首先,提出了一个综合考虑局部密度ρi和相对距离δi的参数μi,根据μi的排列顺序及下降趋势trend自动确定聚类中心.然后,基于基尼系数G对截断距离dc做了自适应选择.最后,对ADPC算法做出了实验验证,并与DPC算法和K-means算法进行了对比.实验结果表明,ADPC算法具有较高的ARI,NMI和AC值,具有较好的聚类效果.

关键词: 聚类;自适应;聚类中心;截断距离;下降趋势;基尼系数

Abstract: The density peak clustering algorithm cannot adaptively cluster because it cannot adaptively select the clustering center and cutoff distance dc according to data set, so that an adaptive density peak clustering(ADPC) algorithm was proposed. Firstly, a parameter μi that comprehensively considers the local density ρi and cutoff distance δi was proposed, and the cluster center was automatically determined according to the sorting and downtrend of μi. Then, an adaptive selection of dc was made based on the concept of Gini coefficient. Finally, the ADPC algorithm was verified and compared with the DPC and K-means algorithm. The experimental results show that the ADPC algorithm has higher ARI, NMI and AC values, and has a better clustering effect.

Key words: clustering; adaptive; clustering center; cutoff distance; downward trend; Gini coefficient

中图分类号: