Abstract:A meta generalization based Bagged Boosting algorithm is proposed to obtain high prediction precision learner, on the basis of the conventional Boosting algorithm which concerns training precision only. This algorithm modifies the step length during the Boosting optimization procedure according to the generalization error estimation of the weak learner. Thus it assigns the weight to each weak,learner proportional to its generalization capability. A new Bagging algorithm, with the characteristic of "half sampling without replacement", is applied to construct the weak learner, which leads to less computation burden and more convenience in the generalization error estimation, compared with the classic one.

A Meta Generalization Based Bagged Boosting Algorithm