电子学报 ›› 2021, Vol. 49 ›› Issue (8): 1524-1532.DOI: 10.12263/DZXB.20200626

• 学术论文 • 上一篇    下一篇

基于阴影集的多粒度三支聚类集成

姜春茂, 赵书宝   

  1. 哈尔滨师范大学计算机科学与信息工程学院,黑龙江 哈尔滨 150025
  • 收稿日期:2020-06-29 修回日期:2020-12-21 出版日期:2021-08-25 发布日期:2021-08-25
  • 作者简介:姜春茂 男,1972年生于黑龙江省哈尔滨市,工学博士,哈尔滨师范大学教授,硕士生导师,主要研究方向为云计算、嵌入式计算、三支决策理论. E-mail:hsdrose@126.com
    赵书宝(通讯作者) 男,1996年生于河南省周口市,硕士研究生,主要研究方向为云计算、机器学习、数据挖掘、三支决策理论. E-mail:machinelearner@126.com
  • 基金资助:
    黑龙江省自然科学基金(LH2020F031)

Multi-granulation Three-Way Clustering Ensemble Based on Shadowed Sets

Chun-mao JIANG, Shu-bao ZHAO   

  1. College of Computer Science and Information Engineering,Harbin Normal University,Harbin,Heilongjiang 150025,China
  • Received:2020-06-29 Revised:2020-12-21 Online:2021-08-25 Published:2021-08-25

摘要:

聚类集成旨在通过融合多个不同的基聚类结果得到一个统一的类簇划分.针对现实环境中的模糊和不确定性数据,本文提出了一种基于阴影集的多粒度三支聚类集成算法.算法首先使用FCM聚类产生一组有差异性的基聚类成员,并通过阴影集构造三支聚类.然后引入多粒度粗糙集构建了四个近似集合,将每一个类簇划分为一个核心域和三个边界域.最后对边界域中的数据依次划分到核心域中,无法划分的对象则留在边界域,最终得到了三支聚类集成的结果.实验结果表明,本算法在准确率、调整兰德系数和归一化互信息方面,与多种现有的聚类集成算法相比得到了更好的聚类集成结果.

关键词: 聚类集成, 三支聚类, 多粒度, 模糊c均值(FCM), 阴影集

Abstract:

The purpose of clustering ensemble is to find a unified partition of objects by fusing a set of clustering results. This paper proposes a multi-granulation three-way clustering ensemble algorithm based on shadowed sets to deal with the fuzzy and uncertainty data in the actual world. First, the algorithm generates a set of clustering members using the fuzzy c-means algorithm, and then the membership degree is mapped into three regions to construct three-way clustering. Second, the multi-granulation rough sets are used to construct four different approximate regions. Each cluster contains a core region and three boundary regions. Finally, the shadowed set is used to classify objects in boundary regions sequentially. Objects that cannot be divided are left in the boundary region. The experimental results show the algorithm obtains better clustering ensemble results in accuracy, adjust rand index, and normalized mutual information compared to multiple existing clustering ensemble algorithms.

Key words: clustering ensemble, three-way clustering, multi-granulation, fuzzy c-means (FCM), shadowed sets

中图分类号: