A Classification Approach in the Probability of Credit Card Approval using Relief-Based Feature Selection

Credit card approval is important in all financial businesses; an applicant may deny and approve their credit card based on their financial record and personal information. Right decision-making for issuing credit cards for applicants is important, it may lose the financial asset of the companies if may had a bad decision in issuing the credit card. This study used Credit Card Approval in an Open Dataset online repository with 19 attributes and 304, 356 instances, and also used different classification models to predict the best attributes in the dataset. 10-fold-cross validation is used for testing and training the data, and ReliefF-Based feature selection to rank the best attributes in the dataset and shows that 12 attributes and 132, 492 instances remained. As a result, occupation type is the best attribute that may consider in applicants' records, and the flag of their own car is the least. The best classification model in the study is Random Forest gained 95.76%, followed by KNN with 94.37%, and lastly NN has 71.56%. This result is evaluated using a confusion matrix for error rate and ROC curve for handling overfitting in an unbiased result. This approach can be a significant input in a financial institution when implemented in an online pre-evaluation system instead of a manual assessment.