Smoothing spline regression: Scalable computation and cross validation
Abstract
We consider the applicability of smoothing splines via the penalized likelihood method to large data sets. Smoothing spline methods with large data suffer from expensive computations which limit the practical use. This thesis studies fast computations of the spline estimates via lower dimensional approximations in penalized likelihood regression with responses from exponential families and censored survival data in accelerated life models by choosing a random subset of basis functions. We conduct an asymptotic analysis of convergence rates followed by an algorithmic computation and simulation studies. The Bayes model associated with the approximations is discussed to provide interval estimates. We also discuss a simple modification of Generalized Cross Validation method in Gaussian regression and Generalized Approximate Cross Validation method in non Gaussian regression for smoothing parameter selection.
Degree
Ph.D.
Advisors
Gu, Purdue University.
Subject Area
Statistics
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.