Multidimensional Classification with Semiparametric Mixture Model

• Activity Number: 30
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Government Statistics Section
• Author(s): Ao Yuan* and Chunxiao Zhou
• Companies: NIH and NIH
• Keywords: EM algorithm ; multi-dimensional classification ; profile likelihood ; semiparametric model ; social security data

Abstract

It is known that a pre-specified parametric model is efficient if the model is correctly specified, but not robust against miss-specification; and that the nonparametric method is robust but not efficient. In practice often some paremetric model is known to be correct to some extent. We propose and study a semiparametric model in this setting for multi-dimensional classification problem. The model is mixture with a known component and an unknown component, the profile likelihood is used for the model parameter estimate. Then we use the Neyman-Pearson classification rule to classify the subjects according to this semiparameytric model. Large sample properties of this method is investigated, and simulation studies conducted to evaluate the performance of the proposed model, and it will be applied to the classification procedures in the Social Security Administration disability program.