稳健的t-MFA模型的极大似然估计

ML estimate of the robust t-MFA model

混合因子分析模型是一种非线性的分析高维数据的工具.但是,当一组观测数据中包含比正态尾部长的一些数据点时,其模型的拟合效果受到严重影响.为此,结合稳健的多元t分布提出基于t分布的混合因子分析模型,同时,在AECM算法基础上提出适合此模型的参数估计方法.最后,通过模拟实验,说明模型的优越性.

Mixtures of factor analyzers,is a nonlinear tool for high-dimension data. However, for a set of observation data containing a group or groups of data longer than normal tails,the use of model from the doeuments[1,2] may unduly affect the fit performance. This paper presents a mixture of factor analyzers based on the robust multivariate t distribution; mean while, based on the AECM algorithm, the suitable method to fit this t mixture of factor analyzers is discussed. In the end,an artificial numerical example is given to illustrate our results.