Double-bagging: combining classifiers by bootstrap aggregation

摘要: The combination of classifiers leads to substantial reduction misclassification error in a wide range applications and benchmark problems. We suggest using an out-of-bag sample for combining different classifiers. In our setup, linear discriminant analysis is performed the observations sample, corresponding variables computed bootstrap are used as additional predictors classification tree. Two combined therefore method variable selection bias no problem estimate error, need test disappears. Moreover, procedure performs comparable best number artificial examples applications.