A new algorithm for imbalanced datasets in presence of outliers and noise
Shujuan Zhao, Hua-Peng Zhang, Lei Li · 2012
In recent years, learning from imbalanced datasets has attracted much attention both in academic and industrial fields. The kernel modification method based on Riemannian metric is an effective method to handle the class imbalance problem. But it cannot deal with the outliers and noise in the imbalanced datasets. However, Fuzzy Support Vector Machine (FSVM) can deal with the outliers and noise in the balanced datasets. In this paper, we combine the FSVM with the kernel modification method based on Riemannian metric together to handle the imbalanced datasets in presence of outliers and noise. Experimental results on four UCI datasets show this method to be effective in improving class prediction accuracy.