Combining SVMS for Classification on Class Imbalanced Data

The class imbalance problem in classification scenarios is considered to be one of the main issues that limits the performance of many learning techniques. When reporting high classification accuracy a classifier may still exhibit poor performance for the minority class that is often the class of interest. In this paper, we propose to address the class imbalance problem by applying an SVM-based ensemble framework that provides the ability to control the trade-off between discovery rate of the under-represented classes and the overall accuracy simultaneously. We evaluate the performance of the proposed technique on both synthetic and real-world datasets demonstrating the advantage of the method compared to state-of-the-art approaches.