A Comparison of Credit Card Approval Prediction Efficiency between Four Classification Methods of Thai Commercial Banks

Abstract

In this study, credit card approval prediction results of 4 classification methods were determined and tested against real data obtained from two Thai commercial banks, and their prediction efficiencies were compared. The tested classification methods were the following: 1) k-nearest neighbor method using IBk algorithm; 2) decision tree method using J48 algorithm; 3) neural network method using multilayer perceptron algorithm; and 4) support vector machine method using polynomial kernel. The following performance values were employed: accuracy, true positive rate, true negative rate, precision, recall, F-Measure, false positive rate, false negative rate, mean absolute error (MAE), and mean square error (MSE).

The important results are as follows. The neural network method showed the best accuracy, true positive rate, false negative rate, recall, F-Measure, MAE, and MSE at 70.37%, 0.821, 0.179, 0.821, 0.742, 0.3683 and 0.1925 respectively. The decision tree method and k-nearest neighbor method exhibited the best true negative rate and false positive rate at 0.731 and 0.269 respectively, while the best precision at 0.682 was attained by the decision tree method. Since the neural network method offered the best efficiencies for 7 out of the 10 values, it was considered the best prediction method for credit card approval