计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (16): 10-17.DOI: 10.3778/j.issn.1002-8331.1704-0227

• 热点与综述 • 上一篇    下一篇

基于密度比例的密度峰值聚类算法

高诗莹1,2,3,周晓锋2,3,李  帅2,3   

  1. 1.东北大学 计算机科学与工程学院,沈阳 110000
    2.中国科学院 沈阳自动化研究所,沈阳 110016
    3.中国科学院 网络化控制系统重点实验室,沈阳 110016
  • 出版日期:2017-08-15 发布日期:2017-08-31

Clustering by fast search and find of density peaks based on density-raito

GAO Shiying1,2,3, ZHOU Xiaofeng2,3, LI Shuai2,3   

  1. 1.School of Computer Science and Engineering, Northeastern University, Shenyang 110000, China
    2.Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China
    3.Key Laboratory of Control Network System, Chinese Academy of Sciences, Shenyang 110016, China
  • Online:2017-08-15 Published:2017-08-31

摘要: CFSFDP(Clustering by Fast Search and Find of Density Peaks)是一种新的基于密度的聚类算法。该算法可以对非球形分布的数据聚类,有待调节参数少、聚类速度快等优点。但是对于类簇间密度相差较大的数据,该算法容易遗漏密度较小的类簇而影响聚类的准确率。针对这一问题,提出了基于密度比例峰值聚类算法即R-CFSFDP。该算法将密度比例引入到CFSFDP中,通过计算样本数据的密度比峰值来提高数据中密度较小类簇的辨识度,进而提升整体聚类的准确率。基于9个常用测试数据集(2个人工合成数据集,7个UCI数据集)的聚类实验结果表明,对于类簇间密度相差较大和类簇形状复杂的数据聚类问题,R-CFSFDP能够使得类簇中心更加清晰、易确定,聚类结果更好。

关键词: 聚类, 密度峰值, 密度比例, 密度变化

Abstract: CFSFDP(Clustering by Fast Search and Find of Density Peaks)is a new density-based clustering algorithm, which can cluster the non-spherical data with fewer parameters and high speed of clustering. However, when the density of different clusters vary widely, it is hard to find the clusters with sparse density, so that the accuracy of clustering will be decreased. To solve this problem, this paper proposes a density-raito based CFSFDP that short of R-CFSFDP. In this algorithm, the density-ratio is introduced into CFSFDP to make clusters with sparse density easily identifiable. To validate the algorithm, experiments are conducted with 9 data sets (2 synthetic data sets, 7 UCI data sets). The experimental results show that, when the cluster shape is complex and the density of different clustersvary widely, it makes the cluster centers easier to be determined and has a higher accuracy of the clustering than CFSFDP.

Key words: clustering, density peaks, density-raito, varying densities