B1489
Title: Sparse estimation in heterogeneous varying coefficient regression models
Authors: Abbas Khalili - McGill University (Canada) [presenting]
Abstract: Statistical methodologies are presented based on the regularized local-kernel likelihood for parameter estimation and feature selection in a sparse finite mixture of varying coefficient regressions. These models are commonly used to learn heterogeneous effects of features on a response variable where there is unobservable heterogeneity in data, and features' effects also vary according to an index variable such as time or location. Although complex, this situation frequently occurs in real data applications, which we demonstrate using a genetic data set. We will discuss the large-sample properties of the proposed methods, and we also evaluate their finite-sample performance via simulations. Finally, we will discuss the results of our real data analysis.
