Data mining in retail credit

Abstract

This article presents a real-world application of a data mining approach to credit scoring. It describes the development and the validation of a decision tree, which aims to discriminate between good and bad accounts of Littlewoods Home Shopping customers based on a sample of orders placed between January and November of 2000. This decision tree was constructed for the orders referred to the Authorisations Department. It showed a great improvement in performance compared to the current manual decisions taken for the orders referred to this Department. The implementation of this tree indicates that Authorisation Advisors should apply a set of simple rules in order to optimise their decision making process. The methodology of the decision tree construction is presented in detail. Furthermore the article discusses alternative approaches to credit scoring. Logistic regression is the most widely used technique and it can be used as a benchmarking to assess competing approaches in credit scoring. Using the Receiver Operating Characteristic (ROC) curve as a performance measure of predictive accuracy, the superiority of the decision tree model against the logistic regression model is indicated.