Modified blame-based noise reduction for concept drift

Competence enhancement plays an important role in case-base editing. Traditional competence enhancement methods tend to omit the evolving nature of a case-based learner, but take the whole case-base as a static training set. This may seriously delay or even prohibit a learner from learning new concepts, when concept drifts. This paper proposes a Modified Blame Based Noise Removal algorithm (M-BBNR). Our MBBNR algorithm preserves some potential noise cases, in case of representing novel concepts. Experiment show that with such a "wait-and-see" policy, the developed M-BBNR algorithm outperforms other famous competence enhancement methods on real world dataset and is able to tuning the case-base according to the concept drift effectively.