Abstract:
|
We propose a novel approach to building regression trees and ensemble learning in survival analysis. By first extending the theory of censoring unbiased transformations, we are able to construct observed data estimators of full data loss functions with right-censored outcomes. This theory provides the basis for constructing several new algorithms for building regression trees and regression ensembles. For the particular case of squared error loss, we show how to implement these algorithms using existing software (e.g., CART, random forests) by making use of a related form of response imputation. Comparisons of these methods to existing ensemble procedures for predicting survival probabilities are provided in both simulated settings and through applications to several datasets. If time permits, extensions to the setting of competing risks are then considered. The contents of this presentation represent joint work with Jon Steingrimsson, Liqun Diao, Youngjoo Cho and Annette Molinaro.
|