Skip to Main Content
This paper investigates the credit scoring accuracy of five data mining technologies for bank credit cards: C5.0 decision tree, neural network, chi-squared automatic interaction detector, stepwise logistic model and classification and regression tree. Firstly, we extract a comprehensive variable from the raw data by using principle component analysis to indicate the customers' default or not. Then we build the credit scoring models using data mining technologies and compare forecasting effects of the five models. Finally, we discuss how to classify non-defaulting applicants by using stepwise logistic model extensively.
Date of Conference: 3-4 Dec. 2011