高维异方差高斯混合罚模型聚类

High-dimensional Heteroskedasticity Penalized Gaussian Mixture Model-Based Clustering

罚模型聚类实现了在聚类过程中精简变量的目标,同时如何识别聚类的有效变量成了一个新的问题.在这个问题上,已有的研究有成对罚模型,模型处理了各类数据同方差的情况.考察了异方差情况下的变量选择问题,针对异方差数据提出了两种新的模型,并给出模型的解和算法.模拟数据分析结果表明,异方差数据上两个新模型都有更好的表现.

Penalized model based clustering achieved the goal of deleting variables in the process of clustering, while how to recognize the clustering function of those undeleted variables becoming a new problem. Researches on this problem include the pairwise penalized model which dealt with the situation of homoscedasticity data. This paper investigated the problem of variables selection under the heteroscedasticity situation, and proposed two new models and their solution ＆ algorithm respectively for the heteroscedasticity data. Simulation data analysis indicate that both two models perform better on the heteroscedasticity data.