计算机工程与应用 ›› 2011, Vol. 47 ›› Issue (29): 175-181.

• 图形、图像、模式识别 • 上一篇    下一篇

推进式优化特征权重的K-中心点聚类方法

陈新泉1,2   

  1. 1.重庆三峡学院 计算机科学与工程学院,重庆 404000
    2.上饶师范学院 数学与计算机科学学院,江西 上饶 334001
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2011-10-11 发布日期:2011-10-11

Boosting K-medoids clustering method based on feature weight optimization

CHEN Xinquan1,2   

  1. 1.School of Computer Science and Engineering,Chongqing Three Gorges University,Chongqing 404000,China
    2.School of Mathematics and Computer Science,Shangrao Normal University,Shangrao,Jiangxi 334001,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-10-11 Published:2011-10-11

摘要: 为获得更贴近于混合属性数据点集空间的相异性度量,从而探测出数据点集的更有意义的聚类分布,提出了一种推进式优化特征权重的K-中心点聚类算法。对该聚类算法进行了必要的讨论,给出其时间复杂度分析及算法收敛性分析。为实现该聚类算法的特征权重优化步骤,给出了二种不同的特征权重优化方法和几个自适应优化距离权重系数、目标函数系数的方法。这些优化方法在一定的理论层次上解决了相异性度量的自适应优化问题。通过几个UCI标准数据集验证了该聚类算法有时能取得更好的聚类质量,从而说明该加权聚类算法具有一定的有效性。给出了几点研究展望,为下一步的研究指明了方向。

关键词: 相异性度量, K-中心点聚类, 有序属性, 无序属性, 混合属性

Abstract: In order to search a better dissimilarity measure of one hybrid attributes space and find more meaningful clustering distributions of some data sets,this paper presents a boosting K-medoids clustering algorithm based on feature weight optimization.After some necessary discuss,it lists the time complexity and astringency analysis of the boosting K-medoids clustering algorithm.In order to implement the feature weighting part of this clustering algorithm,it gives two different methods of optimizing feature weight and several adaptive methods of optimizing distance weighting coefficient and objective function coefficient.These optimization methods have solved the adaptive optimization problem of dissimilarity measure in some theoretical level.The new clustering algorithm can often get a better clustering quality which is validated by experiments of several UCI standard data sets.It indicates several valuable research expectations.

Key words: dissimilarity measure, K-medoids clustering, order attributes, sorted attributes, hybrid attributes