Predicting Probability of Credit Card Default at the Stage of Credit Card Application Using Supervised Machine Learning Approaches

Abstract

Increasing Non-Performing Loans and advances of Banking and Financial institutions have become one of the main problems in this era. Specially, increasing the number of willful defaulters of the banking and finance sector. Credit Card portfolio is representing the major part of loans and advances portfolio of a Bank. Banks are issuing credit cards by analyzing the credit card application. Except for fully secured credit cards, other credit cards are not required to attach any security for recovering the non-performing amount. Therefore, the Managers and the officers who are issuing credit cards must have better understanding and experience to analyze the credit card applications received by customers. But, most of analysis techniques using by them have more personal judgements rather than analyzing it in a proper manner. Therefore, proper credit card application analysis and customer credit scoring model is need for this type of transactions. This research study is focusing to develop a machine-learning process using supervised learning methods to predict Credit Card default probability of Credit Card applicants before issuing the Credit Card to customers. This model can apply to the banking sector of Sri Lanka to reduce ability of transferring Credit Cards to Non-Performing section. The review of the literature guided to identify the previous studied in relating to previously developed models to predict of transferring Non-Performing Credit Cards research problem. To develop a supervised learning model and testing the model by using the data in relating to Bank of Ceylon credit card client’s is used for this study. The research study will be conducted using supervised learning algorithms Naïve Bayes, decision trees, linear regression, logistic regression, k-nearest neighbor algorithm and support vector machines. After analyzing, using all of these algorithms finally the best algorithm for this kind of prediction will be selected.