Right-Censored Nonparametric Regression: A Comparative Simulation Study

Abstract

This paper introduces the operating of the selection criteria for right-censored nonparametric regression using smoothing spline. In order to transform the response variable into a variable that contains the right-censorship, we used the Kaplan-Meier weights proposed by [1], and [2]. The major problem in smoothing spline method is to determine a smoothing parameter to obtain nonparametric estimates of the regression function. In this study, the mentioned parameter is chosen based on censored data by means of the criteria such as improved Akaike information criterion (AICc), Bayesian (or Schwarz) information criterion (BIC) and generalized cross-validation (GCV). For this purpose, a Monte-Carlo simulation study is carried out to illustrate which selection criterion gives the best estimation for censored data.