A SURVEY OF ENSEMBLE CLASSIFICATION OVER CONCEPT DRIFT DATA STREAMS

Ensemble classifier is one of the most common methods used to handle concept drift. The important technologies of ensemble classification algorithms are investigated in this paper. Firstly, the types of concept drift are introduced, which include sudden, gradual, recurring and incremental. And ensemble classification algorithms to dealing with different types of concept drift are summarized. Secondly, the classic and the latest status of ensemble classification algorithms based on the learning strategies, such as bagging, boosting and base classifier combination are analyzed synthetically. And the main techniques, advantages and disadvantages of the algorithms are summarized. Finally, the future work of ensemble classification over data streams are discussed.