J4 ›› 2012, Vol. 39 ›› Issue (5): 61-65+78.doi: 10.3969/j.issn.1001-2400.2012.05.011

• 研究论文 • 上一篇    下一篇

一种改进的粒子群优化快速聚类算法

王纵虎;刘志镜;陈东辉   

  1. (西安电子科技大学 计算机学院,陕西 西安  710071)
  • 收稿日期:2011-08-04 出版日期:2012-10-20 发布日期:2012-12-13
  • 作者简介:王纵虎(1984-),男,西安电子科技大学博士研究生, E-mail:zonghuwang@gmail.com.
  • 基金资助:

    国家科技支撑计划资助项目(2012BAH01F05);国家自然科学资金资助项目(61173091)

Improved PSO-based fast clustering algorithm

WANG Zonghu;LIU Zhijing;CHEN Donghui   

  1. (School of Computer Science and Technology, Xidian Univ., Xi'an  710071, China)
  • Received:2011-08-04 Online:2012-10-20 Published:2012-12-13

摘要:

提出了一种改进的基于粒子群优化的快速K均值算法,有效克服了K均值算法对初始聚类中心敏感和容易陷入局部最优从而影响聚类效果等缺点.与已有的粒子群优化聚类算法相比,该算法通过对样本各维属性进行规范化,预先计算样本的相异度矩阵,提出了一种简化的粒子的编码规则,基于相异度矩阵进行粒子群优化K均值聚类, 在保证聚类效果的基础上,有效降低了计算的复杂度.在多个UCI数据集上的实验结果表明,该算法是有效的.

关键词: 粒子群优化, 聚类, K均值, 相异度, 适应度

Abstract:

This paper presents an improved particle swarm optimization based fast K-means algorithm which effectively overcomes the shortcomings of the K-means algorithm such as sensitive to initial cluster centroid and easiness to fall into local optimum so as to affect the clustering results. Compared with the existing particle clustering algorithm, is algorithm first normalizes the attributes of all the samples, and then computes the dissimilarity matrix. We propose simplified particle encoding rules and use PSO-based K-means clustering based on the dissimilarity matrix to ensure the basis for the clustering effect and reduce computational complexity. Experimental results on several UCI data sets validate the advantages of the proposed algorithm.

Key words: PSO, clustering, K-Means, dissimilarity, fitness

中图分类号: 

  • TP391