确定最佳聚类数的二阶差分统计法  

A second-order differential statistical method for determining the optimal cluster number

作　　者：崔建斌[1] 姬安召[2] 张科[3] 

机构地区：[1]陇东学院数学与统计学院,甘肃庆阳745000 [2]陇东学院能源工程学院,甘肃庆阳745000 [3]西北师范大学数学与统计学院,甘肃兰州730070

出　　处：《安徽大学学报（自然科学版）》2017年第2期61-67,共7页Journal of Anhui University(Natural Science Edition)

基　　金：甘肃省科技计划基金资助项目(1606RJZM092;1506RJYM324)

摘　　要：多元统计数据的聚类是分析数据的一类非常重要方法,但是如何确定最佳聚类数往往是比较困难的.针对分析多元统计数据中应用最广泛的方法之一——K均值聚类方法,提出了一种新的二阶差分统计确定最佳聚类数的方法,并将其与CH统计法、Gap统计法进行了对比.实证分析表明,论文所提方法更为简单而且有效.Clustering in multivariate statistics is a very important method for data analysis. But how to determine the optimal cluster number is usually difficult. K-mean cluster is one of the most widely used method in multivariate data analysis. Based on the K-mean cluster, a new second-order differential statistical methods was proposed to determine the optimal cluster number. It was also compared with CH statistical method and Gap statistical method. Results indicated that the proposed second-order differential statistical method was more simpler and more effective.

关 键 词：二阶差分 聚类 最佳聚类数 仿真实验 

分 类 号：O212.4[理学—概率论与数理统计]