All Times ET
Keywords: cross-validation, half-sampling, model selection, variance estimation
This talk concerns the problem of variance estimation of cross-validation. We consider the unbiased cross-validation risk estimate in the form of a general U-statistic and investigate how to assess the variation of the risk estimation. We propose an efficient variance estimator under a half-sampling design that is proved to be first-order unbiased. Furthermore, we discuss a practical approach to estimate its bias using the internal variance estimation method to obtain a bias-corrected variance estimator. The numerical results suggest that the proposal is comparable or better than its competitors in achieving a smaller mean squared error with a negligible bias, even when the proposal is realized without bias correction. What is more, the developed variance estimator is much more computationally efficient than bootstrap and jackknife.