A1149
Title: Sparse estimation in semi-parametric finite mixture of regression models subject to right censoring
Authors: Farhad Shokoohi - University of Nevada Las Vegas (United States) [presenting]
Abstract: The aim is to propose a sparse estimation framework for semi-parametric finite mixture of regression models in the presence of right-censored outcomes. Such models are particularly relevant for heterogeneous populations where subgroups exhibit distinct regression relationships, yet the underlying error distributions are left unspecified. The approach integrates a penalized likelihood method with sieve-based nonparametric density estimation to accommodate both component heterogeneity and censoring. The sparsity-inducing penalty facilitates variable selection within each mixture component, enhancing interpretability while maintaining predictive accuracy. The asymptotic properties of the proposed estimator are established, including selection consistency and oracle efficiency, under mild regularity conditions. Extensive simulation studies demonstrate superior performance in identifying relevant covariates and recovering component structures compared to existing methods. An application to a real survival dataset illustrates the practical utility of the method in uncovering latent subpopulation-specific effects in censored regression settings.
