基于隐变量的有限混合模型对有序数据的Bayes聚类分析  

Bayesian Clustering for Ordinal Data Based on Finite Mixture Models of Latent Variables

作　　者：徐勤丰[1] 俞燕[1] 孙鹏飞[1] 

机构地区：[1]复旦大学统计学系,上海200433

出　　处：《应用概率统计》2007年第4期407-418,共12页Chinese Journal of Applied Probability and Statistics

基　　金：国家自然科学(10271078).

摘　　要：本文基于隐变量的有限混合模型,提出了一种用于有序数据的Bayes聚类方法.我们采用EM算法获得模型参数的估计,用BIC准则确定类数,用类似于Bayes判别的方法对各观测分类.模拟研究结果表明。本文提出的方法有较好的聚类效果,对于中等规模的数据集,计算量是可以接受的.Based on finite mixture models of latent variables, we propose a Bayesian clustering metnod for ordinal data. EM algorithm is employed to compute the estimates of model parameters, BIC criterion is adopted to determine the number of clusters, and an analogue procedure of Bayesian discrimination is used to classify each observation. The results of a simulation study show that the method works well, and the computing efficiency is acceptable for a dataset of moderate size.

关 键 词：有序数据 聚类分析 BAYES分析 有限混合模型 EM算法 BIC准则 

分 类 号：O212.4[理学—概率论与数理统计] O212.8[理学—数学]