Adaptive Classification Method for Concept Drift Based on Online Ensemble

In view of the problems that the online learning model cannot respond in time to the change of data distribution and it is difficult to extract the latest information of data distribution after concept drift occurs in streaming data, which leads to slow convergence of the learning model, an adaptive classification method for concept drift based on online ensemble (AC_OE) is presented. On the one hand, the online ensemble strategy is used to construct a local online learner, which can dynamically adjust the weight of base learner by local prediction of training samples in data blocks. It is helpful to not only extract the evolution information of streaming data in depth to make a more accurate response to the change of data distribution, but also improve the adaptability of the online learning model to the new data distribution after the occurrence of concept drift, and the real-time generalization performance of the learning model is improved too. On the other hand, the incremental learning strategy is used to construct a global incremental learner, and incremental training updates are carried out with the entry of new samples. The method extracts global distribution information of streaming data, and the model can maintain good robustness in the steady state of streaming data. Experimental results show that the proposed method can respond to concept drift and accelerate the convergence of online learning model, and improve the overall generalization performance of the learner effectively.