An Ensemble Data Stream Mining Algorithm for Class-Imbalanced Applications

For the category-imbalanced applications, traditional ensemble data stream mining algorithms will result in low accuracy for the small classes and fail to meet the needs of applications. This paper provides a novel class-imbalanced data learning method based on MAE named CIMAE to solve the above problem. Instead of directly using each incoming data, it acquires data blocks for online training each time by setting up a sample library and a sliding window. Compared with traditional data stream mining algorithms, the results showed that CIMAE achieves the state-the-of-art performance for class-imbalanced application.