Title: Identifying the subpopulation-specific covariates in FMR model
Authors: Mengque Liu - The School of Economics, Xiamen University (China) [presenting]
Abstract: A finite mixture of regression (FMR) models for high dimensional inhomogeneous data is considered, where the number of covariates may be much larger than sample size. However, there lack the mechanism to analyze the sub-population characterisation. We propose an $l_0$ norm penalty which is the first to identify the subpopulation-specific important covariates in FMR models. Computationally it is realized using an efficient EM algorithm. Theoretically it has the much desired consistency properties, its oracle results are also provided. Simulation study under diverse settings shows the superior performance of the proposed method. In both simulations and real data analysis, we demonstrate a significant gain in identification rate over the FMRLssso method.