Data augmentation based estimation for the censored composite quantile regression neural network model

Abstract

Composite quantile regression neural network (CQRNN) model has been widely applied to explore complex patterns among variables, but few researchers consider its possible applications in censoring problems (left censoring, right censoring, and interval censoring might occur in the responses y). In this paper, we propose an iterative estimation method based on the data augmentation algorithm for censored CQRNN model. Firstly the censored data are imputed through a data augmentation process, then we update the CQRNN model with the imputed data, finally the updated CQRNN model is employed to make predictions. Simulation studies and real data application illustrate that the proposed method outperforms the existing censored methods in terms of mean absolute error and root mean squared error, meanwhile producing very close results to those of the uncensoring case. The proposed method can be easily adapted to deal with different censoring types including left censoring, right censoring, and interval censoring, remedying the defect that the available censored methods are only suitable for right censoring type.