两高斯混合分布的局部影响分析(英文)

Local Influence Analysis on Two-Component Gaussian Mixture

对于两高斯混合分布,很难求参数的极大似然估计,当然也不便于影响分析.为此,这里通过引入遗失数据并利用EM算法,将局部影响分析建立在完全数据的似然函数基础上,并分析了一种特殊扰动对几个有代表性的参数的估计的影响、且以一例示之.

For the two- component Gaussian mixture, it is hard to work out the maximum likelihood estimates （MLEs） and hence difficulty in the influence analysis. Together with augmenting missing data, EM algorithm is employed in this paper to develop the local influence analysis based on the conditional expectation of the complete data log - likelihood function. The influence of a special kind of perturbation on the estimates of some representative parameters in also analyzed and illustrated by an example.