Ensemble method uses the principle of majority voting. If you have M weak classifiers whose trainings concentrate more on different subsets of the training set by weighting, the overall classifier which combines the result of the classifiers would possibly do better than a single classifier.
Dear Chinedu Pascl thanks for prompt response. kindly some detail please:
Actually iam using dependent ensemble which dont have multiple classifier. its only do ensemble a meta classifier with other one classifier like DT etc. how they both combined reduced classification errors.