Abstract
The Lasso is a very popular model selection tool (Tibshirani, 1996). It is a
regularized linear regression method what shrinks the estimated coefficients and
selects variables at the same time. Although the Lasso is widely used, the criteria
of model selection are still under development. In this work, we review
and compare the performance between several existing model selection criteria,
including Cross-validation by Tibshirani (1996), modified Cp, AIC and BIC by
Zou et al. (2007), Covariance test by Lockhart et al. (2014) and Exact Inference
test by Lee et al. (2014). Also, we propose an alternative criterion and provide
some solutions to problems in the Covariance test and the Lee's test.
Post a Comment