Artificial Neural Network Based Credit Limit Prediction System for Credit Cards

Abstract

The effectiveness of artificial neural network based models in predicting credit limit, based on customer specific parameters for Bank of Ceylon Credit Card Centre, is presented. The implemented networks were based on feed-forward back-propagation technique. This will facilitate to automate the manual process of granting the credit limit to customers, in a web based system. To obtain a higher rank among other banks, the bank has to go with new technology and reduce paper work. This centralized system is aimed to eliminate the paper based work, provide various statistical analyses and amend whole customer portfolio to make quicker, better decisions. A sample of 4000 customer details was used to train the network while 780 data were used to test the effectiveness of the models. One model was developed to predict the exact limit of a neural network while other model was developed including fuzzy techniques to predict credit limit ranges. First model was able to predict the limits with an accuracy of 64±3% while other model was able to predict the limits with an accuracy of 78±3%. The developed web based system is to feed all credit card application details in to the system and then predict the credit limit based on that data and past financial history of the customer. It uses a client-server model with a connected database to allow multiple users to be connected. Asp .NET technologies are used to implement the web system and Microsoft SQL server is used for data storage. In order to achieve the high quality and user satisfaction with the least cost, time and well planned development process, Hybrid Agile model is used. Evaluation of the neural network model was done using past sample data, while evaluation and testing of the web based system was carried out by the team of business users and IS audit users in the bank. It was identified how far the system achieved look and feel, Security, functionality and usefulness. Positive feedback from users proved the system fulfils identified requirements; is easy to use, extremely helpful and beneficial over the manual process.