Predicting Default Payment of Credit Card Users: Applying Data Mining Techniques

Abstract

Over the years, credit card debt crisis is the main issue in share market and card-issuing banks. Most card users, regardless of their payment capability, overused credit cards and cash-card debts. This catastrophe is the biggest challenge for both card holders and banks. The study aimed at predicting the accuracy of default payment of credit card users using data mining techniques. In this study total of six data mining techniques were applied to the data set of 30,000 individual records collected from the UCI data repository. Then we have compared our regression results with target value of the dataset. According to our test results, linear regression shows the best performance with 80% accuracy and Random Forest regression shows the lowest performance with 63% accuracy. Finally, we have evaluated the performance of each algorithm on overall dataset which was randomly sampled and found the Adaboost showing highest performance with 88% accuracy and Random Forest shows lowest performance with 70% accuracy. The study was implemented using data mining tools such as SPSS and Orange.