A Credit Card Fraud Detection Model Using Machine Learning Methods with a Hybrid of Undersampling and Oversampling for Handling Imbalanced Datasets for High Scores

Abstract

Due to recent advancements in communication technology and e-commerce systems, credit cards are currently the most popular form of payment for online as well as offline purchases. As a consequence, there is substantially more fraud involved in the transaction processes. Data mining has an issue for prediction, and data classification, therefore finding occurrences is essential. Unusual events are challenging to be identified due to their irregularity and casualness, but misclassifying unusual events can lead to large financial losses. In this study, an approach for detecting credit card fraud using machine learning methods, such as K-Nearest Neighbors, random forest, decision trees, logistic regression, and support vector machines, is proposed. The research attempts to look at the effectiveness of the classification models while applying both the Oversampling and Undersampling techniques to find instances of fraud in the dataset for fraudulent activities. The experimental study used two days of credit card transactions made by European cardholders in September 2013. To evaluate the models’ performances, confusion matrix, precision, recall, f1_score, cross-validation score, and ROC_AUC score metrics were used. From different experiments of the tested model, it can be easily observed that the performance of all models was better compared with previous literature thus the KNN was the best in almost all metrics used.