Learning from Imbalanced Data Streams Using Rotation-Based Ensemble Classifiers

In this paper, the problem of learning from imbalanced data streams is considered. To solve this problem, an approach is presented based on the processing of data chunks, which are formed using over-sampling and under-sampling. The final classification output is determined using an ensemble approach, which is supported by the rotation technique to introduce more diversification into the pool of base classifiers and increase the final performance of the system. The proposed approach is called Weighted Ensemble with one-class Classification and Over-sampling and Instance selection (WECOI). It is validated experimentally using several selected benchmarks, and some results are presented and discussed. The paper concludes with a discussion of future research directions.