A Hybrid Recovery Scoring Model Using Multivariate Adaptive Regression Splines and Data Envelopment Analysis

This paper aims to devise a cardholder recovery scoring model which enables card issuers to identify creditworthy debtors recoverable from delinquency without misclassification risks. Taking advantage of integrating such highly performed classifiers as artificial neural networks (ANNs) and multivariate adaptive regression splines (MARS) with a relative efficiency evaluation tool, data envelopment analysis (DEA), a classification model with a more desired accuracy is built in the first phase for predicting delinquents´ future credit status, and then DEA model are employed in the second phase to verify the preceding-stage predicted results as well as gain managerial implications on the inefficient delinquents for improvement in the efficiency of card utilization.