A support vector machine credit card fraud detection model based on high imbalance dataset

Abstract

Credit card transactions are exposed to fraudulent activities owing to their sensitive nature. The illegal activities of the fraudsters have been reported to cost financial institutions a lot of money globally as reported in many notable research works. In the past, several machine learning-based approaches have been proposed for the detection of credit card fraud. However, little attention has been given to classification of fraud in high imbalance dataset. Generally, if a dataset is imbalanced, a learning algorithm will give a bias result in terms of the accuracy resulting in poor performance of the model. This study focuses on using Synthetic Minority Oversampling Technique (SMOTE) to address the class imbalance in the selected credit card dataset. Then, ANOVA-F statistic was applied for the selection of promising features. Both the class imbalance and attribute selection techniques were targeted at improving the SVM-based credit card fraud classification. With the balanced dataset, the study achieved an accuracy of 93.9%, recall of 97.3%, precision of 90.3%, and f1 score of 93.5% respectively. It was observed that the result of the Support Vector VM based credit card fraud detection model with class imbalance is better than that of the standard SVM. The study concluded that the class imbalance addressing and attribute selection techniques used were very promising for the credit card fraud detection tasks.