Data mining for automated assessment of home loan approval

Abstract

Banks receive large numbers of home loan applications from their own customers and others each day. In this study, we investigated the use of data mining to decide whether or not to extend credit, based on two analytical approaches: Naïve Bayes and decision tree. Four independent factors were considered: the loan period, the net income of the applicant, the size of the loan, and other relevant characteristics of the potential borrower. Models were constructed that produce three outcomes: approval, conditional approval, and rejection. The predictive accuracy of the models was compared, to evaluate the effectiveness of the classifiers and a Kappa statistic was applied, to evaluate the degree of accuracy with which the models predicted the final outcome. The decision tree model performed better on both accuracy and the Kappa statistic. This model had an accuracy of 90% and a Kappa of 0.8140, whereas the Naïve Bayes had an accuracy of 65% and Kappa of 0.3694. We therefore recommend the use of decision tree-based models for home loan ranking. Data mining of the applicants history can support the decision-making of financial organizations, and can also help applicants realistically evaluate their own chances of securing a loan.