基于层次的K-均值聚类

介绍一种基于层次的K-均值聚类算法(HKMA)。在统计力学的基础上,对传统K-均值聚类划分矩阵里的元素("隶属"概率)做了形式上的改变,并引入一个调控实际聚类数目的因子。这样,在对同一组数据集进行聚类时,调控因子值不同,结果得到的类数目就不同。用一组二维正态分布的数据集和一组用来测试聚类算法的标准数据集(Iris数)进行测试,结果表明该算法具有层次聚类的性质和较满意的聚类精度。

K-Means Clustering Based on Hiberarchy

A K-means clustering arithmetic based on hiberarchy is presented.On basis of statistical mechanics,partition matrix element(membership probability) in traditional K-means clustering is changed and a lagrange multiplier controlling the clusters number is introduced.Thus,for a given dataset,the result gives different clusters number when the lagrange multiplier is not the same.The method is tested on one synthetic and one real datasets.The result demonstrates hiberarchy feature and precision of arithmetic as expected.