SR-based binary classification in credit scoring

Credit scoring is an important process in every financial institution and bank. Its high accuracy in classifying customers helps decrease the credit risk and increase reliability and profit. In this paper, we propose a binary classification approach that can classify customers who apply for loans. A statistical technique called Stepwise Regression (SR) is used as a pre-process to select important features for the classifier. Artificial Neural Network (ANN) is the classification model type that has been chosen. The concept of confusion matrix combined with business rules has been used for obtaining an appropriate classification model. Based on the experimentation, we find that our method, SR-based Binary Classification, has the accuracy rate of 95.65%, which is higher than using ANN alone (91.30%).