摘要
本文考虑了变量权重和样本类别的影响,建立了马氏距离聚类过程中评估协方差矩阵的迭代法。以Fisher的iris数据为样本,运用欧氏距离一般聚类、主成分聚类、改进前后的马氏距离聚类方法,进行实证分析和比较,结果表明本文所提出的新方法准确率至少提高了6.63%。最后,运用该方法对35个国家的相关指标数据进行聚类分析,确定了各国的卫生保健状况等级。
In this paper we establish an iteration method to estimate the covariance matrix of Mahalanobis distance during the cluster analysis,when weights of variables and categories of samples are taken into account.Fisher＇s iris data are analyzed by the Euclidean distance clustering,principal component clustering,unimproved and improved Mahalanobis distance clustering,with the result that the new method has at least 6.63%better accuracy.At last,utilizing the method we analyze some indexes on the hygiene condition of 35 countries to rank the countries.
出处
《数理统计与管理》
CSSCI
北大核心
2011年第2期240-245,共6页
Journal of Applied Statistics and Management
基金
国家自然科学基金资助项目(40537034)
江苏省研究生培养创新工程(CX10B_295Z)
关键词
协方差矩阵
马氏距离
聚类分析
卫生保健
covariance matrix
Mahalanobis distance
cluster analysis
hygiene condition
