Comparison of Undersampling Methods for Imbalanced Credit Card Fraud Dataset

We use credit cards for daily transactions such as shopping, bill payments, electricity bill payments, recharge of mobiles, set-top boxes, etc. We cannot avoid using cashless payments for an easy life. For online transactions, using credit cards is the most common way. Usually, transactions are legitimate, but there is still little chance of fraud. A class imbalance issue exists as only very few transactions are fraud out of a thousand transactions. We propose a new Attribute-clustering based undersampling algorithm to rebalance the credit card fraud dataset, which is highly imbalanced. The proposed approach enhances fraud detection accuracy for the credit card dataset. For the experiments, a credit card dataset is taken from Kaggle. We have also analyzed five undersampling methods: RUS, ROSE US, SBC, KMUS, and Center_1nn. Performances of all approaches under consideration are compared using seven different classifiers: SVM, Random Forest, RPART, XGBoost, KNN, Naïve Bayes, and LR. We used the F1 score, Precision, Recall, and Accuracy as accuracy metrics for performance comparisons. Experimental results show that the XGBoost is the best classifier with an Accuracy of 99.53% for the proposed method, which is the highest Accuracy among all experiments.