Boosting and Functional LARS for High-Dimensional Nonparametric Regression with Grouped Variables

• Activity Number: 40
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Lifeng Wang*+
• Companies: Michigan State University
• Keywords: Boosting ; functional LARS ; high-dimensional additive model

Abstract

In many high-dimensional regression analysis, information on the structures of the data is available and can be effectively utilized to reduce the modeling uncertainty. In this paper, we discuss how to integrate the sparsity and the grouping structure of the predictor variables into regression to improve the predictive performance. We propose a boosting method and a functional LARS algorithm to perform nonparametric regression and feature selection for high-dimensional group additive models. We investigate the learning theory for the proposed boosting algorithm, and illustrate its ?nite sample performance via both simulated and real data.