基于Chameleon算法和谱平分法的聚类新方法  

A New Clustering Method Based on the Chameleon Algorithm and the Spectral Bisection Method

作　　者：张友[1] 赵凤霞[2] 

机构地区：[1]大连民族学院理学院,辽宁大连116605 [2]秦皇岛职业技术学院信息工程系,河北秦皇岛066100

出　　处：《大连民族学院学报》2010年第1期61-64,共4页Journal of Dalian Nationalities University

摘　　要：在分析传统的聚类算法优越性和存在不足的基础上,基于Chameleon算法和谱平分法的思想提出了一种新的聚类方法。相比传统聚类算法而言此算法克服了如k-means算法、EM算法等传统聚类算法在聚类不为凸的样本空间时容易陷入局部最优的缺点,能在任意形状的样本空间上聚类,且收敛于全局最优解,并且可以降低噪声和离群点的影响,提高了算法的有效性。在UCI数据集和5个特殊的二维数据点组成的数据集上进行了实验,证明了本方法的有效性。With analysis of advantages and disadvantages of traditional clustering algorithms, we proposed a new clustering method based on the Chameleon algorithm and the spectral bisection method. Unlike traditional clustering algorithms, this algorithm overcomes a disadvantage of some traditional clustering algorithms such as the k - means algorithm and the EM algorithm, that is, they are prone to local optimum when clustering on a nonconvex sample space. It is capable of clustering on a sample space of any shape and converging to the global optimal solution. It also reduces the influence of the noise and outliers, increasing its effectiveness. We carried out experiments on the UCI data sets and five data sets of special two - dimensional data points and proved the effectiveness of the method.

关 键 词：聚类算法 CHAMELEON算法 谱平分法 k—mean算法 EM算法 不为凸的样本空间 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]