Credit Card Fraud Detection on Class Imbalance Dataset

Abstract

India is growing day by day and a number of enhancements to banking and finance are performed by the government. In this context, the government is frequently supporting digital payments for large as well as small transactions. However, it increases the transparency in payments but in the same ratio, the financial fraud cases are increasing. Among them, credit card fraud is a very common and frequent fraud in the banking system. However, there are a number of automated systems for credit card fraud detection available, but most of them are suffering from the class imbalance problem. The imbalanced training samples are misleading the Machine Learning (ML) algorithm, which leads to an increase in false alarm rates. In this paper, our aim is to contribute an ML method, which is able to deal with the class imbalance issue. Additionally, accurately identify fraud cases. In this context, first, we discuss the class imbalance issue and its available solutions. Then, adopt two appropriate over-sampling methods for handling the class imbalance i.e. ADASYN and SMOTE. Finally, a Binary Convolutional Neural Network has been implemented to classify the over-sampled dataset to classify transactions into fraud and legitimate. The experimental analysis of the model has been carried out based on the Kaggle dataset. The performance results of the proposed technique in terms of accuracy and Area Under the Precision-Recall Curve (AUPRC) are evaluated. According to the obtained results, we found the proposed methodology is enhancing the results and produce up to 99% accurate results.