Identification of Potential Future Credit Card Defaulters from Non Defaulters using Self Organizing Maps

Abstract

Identifying customers who are more susceptible to defaulting on their credit card loans is a subject of interest to the credit card issuing authorities. In the past, incautious approval of credit card applications with low standards has led to major societal implications. Many have used various supervised machine learning techniques to classify such credit card applicants in hope to choose ideal candidates who would not default on their credit card loan. In spite of shortlisting with the help of strong models, cases of customers defaulting on their credit card loans are still not uncommon. Hence, in this paper, we introduce an approach to identify potential future defaulters using an unsupervised deep learning algorithm known as a self-organizing map (SOM). SOM has been extensively used in anomaly detection and dimensional reduction for many areas. It also provides a good two-dimensional representation and visualization of high dimensional data.