Investigating the effect of data augmentation on the performance of machine learning and deep learning methods in detecting fraudulent credit card transactions

With the growth of e-banking in recent years, the rate of fraud in credit card transactions has increased. Therefore, establishing a fraud detection system for financial institutions is of particular importance. The utilized datasets in the fraud detection context always have the problem of class imbalance. Various methods have been used in previous research to build classification models. In this research, we aim to investigate the effect of the data augmentation method on the performance of conventional machine learning methods and deep learning methods. For this purpose, widely-used machine learning techniques, including decision tree, support vector machine, and random forest, along with two deep neural network models are employed. The results of experiments show that data augmentation leads to an increase in the performance of the random forest in terms of F۱-Measure. It achieves the best performance among the compared methods. Also, the results indicate that in general, the use of data augmentation increases the performance of models in terms of recall but decreases precision. Besides, data augmentation reduces the performance of deep learning methods.

کلیدواژه ها:

Fraud detection ، Classification ، Data augmentation ، Deep learning