|
Abstract:
|
Several penalized likelihood methods have been proposed for variable selection, such as LASSO, adaptive LASSO, SCAD. It is desirable to perform variable selection using the L0 penalty, however it is well known that this is NP hard problem. We propose variable selection using a seamless L0 (SEAL) penalty, which closely approximates the L0 penalty and is computationally fast. We study the theoretical properties of the proposed method and show the SEAL estimator has oracle properties. We compare the proposed method with the existing methods using simulation studies and illustrate their application using the Stanford HIV condon data.
|
