A Re-sampling Method for Class Imbalance Learning with Credit Data

Credit rating is a typical class imbalance problem. Over sampling methods are commonly used for dealing with this problem. This paper presents an improved over sampling approach based on synthetic minority over-sampling technique(SMOTE). First, use sample distribution of the minority class to estimate whether different types of samples are crossed. Then generate synthetic samples by samples in different class when different classes cross seriously. In addition, increase weights of part positive samples, which may be not on borderline. At last, the proposed method is evaluated on a credit data set. The results indicate that it is more effective than other methods for the class imbalance learning.