|
Abstract:
|
This paper deals with the bias correction of cross-validation (CV) criterion. It is well known that CV criterion is the second-order unbiased estimator of the risk function measuring discrepancy between the candidate model and true model. A usual bias correction is generally achieved by subtracting an estimated bias from the target estimator. However, such an estimated bias usually becomes complicated, e.g., it depends on higher-order cumulants and higher-order derivatives. In this paper, we reduce the bias of CV criterion to the third order by subtracting a simple term from CV criterion. Our bias-corrected CV criterion can be obtained without calculating explicit forms of the bias terms in CV criterion.
|
