Variable Selection for Varying-Coefficient Models via the Elastic Net Regularization

• Activity Number: 319
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Hidetoshi Matsui*+ and Toshihiro Misumi
• Companies: Kyushu University and Astellas Pharma Inc.
• Keywords: Elastic net ; Group lasso ; Longitudinal data ; Variable selection ; Varying-coefficient model

Abstract

Longitudinal data analysis has been widely used in various fields such as bioscience, ergonomics and meteorology. A varying-coefficient model is one of the useful tools for analyzing longitudinal data. It can effectively describe the relationship between predictors and responses repeatedly measured. We consider the problem of selecting variables in the varying-coefficient model via the elastic net regularization, one of the sparse regularization. Coefficients given as functions are expressed by basis expansions, and then parameters involved in the model are estimated by the penalized maximum likelihood method with the group elastic net penalty. We apply the coordinate descent algorithm derived for solving the problem of sparse regularization. Furthermore, we derive a model selection criterion for evaluating the varying-coefficient model estimated by the regularization method. We examine the efficiency of our modeling procedure through a real data example.