Regression with right-censored high-dimensional data: An application with different imputation techniques

10.48129/kjs.splml.18961

Abstract

This study aims to introduce two modified linear estimators for the right-censored high-dimensional data. Obviously, data of interest involves two important problems to be solved that are censorship and high dimensionality. The introduced estimators are distinguished from other studies in the literature in that it achieves to handle these two problems simultaneously. To solve the censorship problem, two imputation techniques are considered based on machine learning algorithms kNN and Sliding-windows methods respectively. The high-dimensionality problem is handled by the weighted ridge approach which provides estimator with less risk than its alternatives because it detects the covariates with a weak contribution via the post-selection procedure. In addition, the weighted ridge approach resolves the multicollinearity problem effectively. To show the empirical performance of the introduced estimators, a detailed simulation study is carried out and results are presented.