Adaptive group bridge selection in the semiparametric accelerated failure time model

Abstract

The group bridge penalized method has been studied in the multiple linear regression model and the semiparametric accelerated failure time (AFT) model and demonstrated the capability to remove unimportant groups, however, it cannot effectively remove unimportant variables within the important groups. To overcome this limitation, we propose the adaptive group bridge method in the AFT model. We show that the adaptive group bridge method enjoys the powerful oracle property. Simulation studies indicate that the adaptive group bridge approach for the AFT model can correctly identify both important groups and important within-group individual variables even with high censoring rates in high-dimensional data. The PBC data is analyzed to illustrate the application of the proposed method.