关键词: 数据流，概念漂移，分类，噪音

Abstract: Mining with data stream concept drift is a hot topic in data mining. Existing classification approaches consist of ensemble method based on single base classifiers and ensemble method based on hybrid base classifiers,which depend on the stationary assumption and lcarnable assumption. However, the former probably causes the larger classification deviation and the performance on accuracy is impacted in the noisy data streams,while the latter performs worse on the classification accuracy or the time consumption. Motivated by this, an ensembling classification method WE-DTB was proposed, based on hybrid based models with decision trees and Naive Bayes. It is an extended framework of WE model.Meanwhile, we utilized the popular concept drift detection mechanisms based on Hocffding Bounds and μ test to implement the detection on concept drifts. Extensive experiments demonstrate that our proposed method WE-DTB can detect concept drift effectively while maintaining the good performance on classification accuracy and consumptions on time and space.

Key words: Data streams, Concept drifts, Classification, Noise