GAMSel: A Penalized Regression Approach to Model Selection for Generalized Additive Models

• Activity Number: 178
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Alexandra Chouldechova*+
• Companies: Stanford University
• Keywords: generalized additive models ; penalized regression ; coordinate descent ; group lasso

Abstract

In many applications it may be too restrictive to suppose that the effect of all of the predictors is captured by a simple linear fit. Generalized additive models allow for greater flexibility by modeling the linear predictor of a generalized linear model as a sum of more general functions of each variable. We introduce GAMSel (Generalized Additive Model SELection), a method for fitting generalized additive models in high dimensions that allows the effect of each variable to be estimated as being either 0, linear, or a low-complexity curve, as determined by the data. We present a blockwise coordinate descent procedure for optimizing the penalized likelihood objective.