Ensemble margin resampling approach for a cost sensitive credit scoring problem

In the past few years, a growing demand for credit compel banking institution to contemplate machine learning techniques as an answer to obtain decisions in a reduced time. Different decision support systems were used to detect loans defaulters from good loans. Despite good results obtained by these systems, they still face some problems such as imbalanced class and imbalanced misclassification cost problems. In this work, we propose a cost sensitive credit scoring system, based on a two-step process. The first is a resampling step which handles the imbalance data problem followed by a cost sensitive classification step that recognises potential insolvent loans red in order to reduce financial loss. A resampling algorithm called ensemble margin for imbalanced instance (EM2I) is suggested to manage imbalanced datasets in cost sensitive learning. We compare our algorithm with other techniques from the state of the art and experimental results on German credit dataset demonstrate that EM2I leads to a significant reduction of the misclassification cost.