A New Multi-Stage Feature Selection and Classification Approach: Bank Customer Credit Risk Scoring

Lots of customers information regularly are stored in the databases of banks. These databases can be used to assess the credit risk. Feature selection is a well-known concept to reduce the dimension of such databases. In this paper, a multi-stage feature selection approach is proposed to reduce the dimension of database of an Iranian bank including 50 features. The first stage is devoted to removal of correlated features. The second stage is allocated to select the important features with genetic algorithm. The third stage is proposed to weight the variables using different filtering methods. The fourth stage selects feature through clustering algorithm. Finally, selected features are entered into the K-nearest neighbor (K-NN) and Decision Tree (DT) classification algorithms. The aim of the paper is to predict the likelihood of risk for each customer based on effective and optimum subset of features available from the customers.