Estimating the error rate of a prediction rule: Improvement on cross- validation. (English) Zbl 0543.62079

This paper addresses the following problem: a prediction rule (model) is constructed on the basis of some data and then it is desired to estimate the error (err) of this rule in predicting future observations. The cross-validation, bootstrap, randomized bootstrap, and double bootstrap estimators of err are discussed in detail and analysed. The relations between these estimators are also established. The paper is mainly concerned with the situation when the observations and their predictions are dichotomous and the ”input” variables are independent and identically distributed.
Reviewer: P.Stoica


62M20 Inference from stochastic processes and prediction
Full Text: DOI