Hybrid Dual-Resampling and Cost-Sensitive Classification for Credit Risk Prediction

The class imbalance in financial data sets is prevalent and problematic when evaluating credit risks. This paper proposes a Hybrid dual Resampling and cost-sensitive classification approach by creating heuristically balanced data sets. Given an imbalanced credit data set, a synthetic minority class is generated using a resampling learning technique based on Gaussian mixture modelling from the minority class data. Simultaneously, k-means clustering is applied to the majority class. Then, feature selection is performed using an Extra Tree Ensemble technique. Finally, a cost-sensitive logistic model is estimated and applied to predict the probability of default using the heuristically balanced datasets. The results show that the proposed technique achieves superior performance in comparison with other imbalanced preprocessing approaches.