Credit Card Approval Prediction by Using Machine Learning Techniques

Abstract

This research is focusing on application of machine learning (ML) techniques to predict customer eligibility for a credit card to mitigate possible future credit risk which may affect the bank’s financial stability and credit performance. Credit card is a credit facility given for a customer by banks and finance companies around the globe. The credit facility has a credit risk for the banks and financial companies. The repayments are least assured and it often ends up as a non-performing credit facility (NPL). To mitigate credit risk banks are assessing applicant’s creditworthiness and checking the eligibility before granting a credit facility. The decision is mostly based on traditional credit scoring models and credit worthiness will not always be accurate. This project aims to help banking and financial institutions to identify and interact with creditworthy customers by using predictive models. We used Artificial Neural Network (ANN) and Support Vector Mechanism (SVM) to develop models. Under ANN we have tested models using different sizes of batches, low and high learning rates. Linear SVM and Nonlinear SVM both models used to evaluate the best SVM method. Statistical methods under filter-based feature selection methods applied for feature selection. Model accuracy checked using Mean Absolute Error, Confusion Matrix, Area Under Curve (AUC) for training and test data. We have evaluated three classifiers and we observed that Nonlinear SVM is performed better than ANN and linear SVM. Nonlinear SVM model Accuracy is 0.88, Precision is 0.88, Recall is 0.90 and AUC is 0.89. Accuracy, Precision and Recall values are higher in Nonlinear SVM than ANN and Linear SVM. Recall rate is 0.90 means the model predicts positive class 90% correctly. We also realized that customer behavior might be different from country to country and application of several real banking datasets not limited to customer demographic and socio-cultural but also other credit facility features including COVID-19 impact to be an area of concern for researchers. Furthermore, whether there is a relationship between Nonlinearity in highly imbalanced class problems with SMORTE application is another area of concern for researchers