A1093
Title: Divide and conquer approaches for nonparametric regression and variable selection
Authors: Rong Liu - University of Toledo (United States) [presenting]
Abstract: The increasing size of massive data calls for new statistical methods, particularly in the field of nonparametric regression, which is flexible but often requires significant computational resources. To address the limitations of CPU, memory, and storage, several distributed frameworks have been proposed for statistical estimation and inference. The purpose is to examine the statistical efficiency and asymptotic distribution of the spline estimation for nonparametric generalized additive partially linear models using the divide and conquer (DAC) approach. Additionally, a variable selection method is proposed based on a majority voting procedure. Through simulation studies, strong support is found for the asymptotic theory and it is demonstrated that the DAC approach is computationally efficient without sacrificing accuracy. This approach is employed in two research projects, one focusing on the airline industry and the other on retention.
