Out-of-bag样本的应用研究

Bagging can improve the classification error of a weak learning algorithm to a large extent through combining some instable base classifiers. Out-of-bag samples are naturally generated in the process of bagging. At present, out-of-bag samples haven been applied to many fields in ensemble learning, such as estimating the generalization error of a bagging ensemble, constructing ensemble classifiers and so on. This paper gives an overview of the applications of out-of-bag samples whose contents and main characteristics are illustrated. Meanwhile, the future research directions of out-of-bag samples are discussed.