Data Augmentation using Generative models for Credit Card Fraud Detection

Credit card transactions have become the preferred mode of payments in developed countries and its utility is rapidly growing in developing countries making frauds an increasingly consequential problem leading to financial losses and erosion of consumer confidence. Although, credit card data is highly class imbalanced and this makes training of models to classify fraud data difficult. This study employs the use of multiple adversarial networks to generate pseudo data to enhance model performance. This study uses the vanilla implementation, Least Squares, Wasserstein, Margin Adaptive, Relaxed Wasserstein of GANs. The distribution of the generated data against original fraud data, the classifier accuracy, convergence for each model and an optimal number of data generations is analyzed. The generated data is then augmented and tested using an Artificial Neural Network model and a 12.86 % increase in recall for a dataset with a class imbalance of initial 579 to 1 is recorded.

[1]  Nan Yang,et al.  Relaxed Wasserstein with Applications to GANs , 2017, ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[2]  Léon Bottou,et al.  Wasserstein GAN , 2017, ArXiv.

[3]  S. Siva Prakash,et al.  Credit Card Fraud Detection using Adaboost and Majority Voting , 2019 .

[4]  Yoshua Bengio,et al.  Generative Adversarial Nets , 2014, NIPS.

[5]  Zhen Wang,et al.  Multi-class Generative Adversarial Networks with the L2 Loss Function , 2016, ArXiv.

[6]  Changjun Jiang,et al.  Random forest for credit card fraud detection , 2018, 2018 IEEE 15th International Conference on Networking, Sensing and Control (ICNSC).

[7]  Yiannis Demiris,et al.  MAGAN: Margin Adaptation for Generative Adversarial Networks , 2017, ArXiv.

[8]  Peter Beling,et al.  Deep learning detecting fraud in credit card transactions , 2018, 2018 Systems and Information Engineering Design Symposium (SIEDS).