Debit Card Fraud Recognition Using Enhanced Auxiliary Classifier Generative Adversarial Networks

Abstract

With the recent increasing trend of fraudulent transactions involving debit cards in Indonesia, fraud recognition for debit card transactions is an important and challenging problem to be examined.The purpose of this research is to recognize fraudulent transactions on debit cards with development of the Enhanced Auxiliary Classifier-Generative Adversarial Network (EAC-GAN) model which is a development of Auxiliary Classifier-Generative Adversarial Network (AC-GAN) model.EAC-GAN uses AC-GAN, Synthetic Minority Over-sampling Technique (SMOTE), Principal Component Analysis (PCA), and tuning parameter to recognize fraud transactions in debit cards and increase F1-Score.SMOTE is used to overcome imbalanced data in debit card transactions dataset.Then PCA is required to reduce dimension of the dataset and to know which factors are influential in explaining the phenomenon in the dataset while maintaining characteristics of the data.Parameter tuning is useful to achieve the best F1-Score in training and testing the EAC-GAN model.This research also explores the study of AC-GAN and Convolutional Neural Network 2 Dimension (CNN2D) performance.The result of this research describes that EAC-GAN model beats CNN2D done in the previous research.F1-Score for EAC-GAN is 74% and F1-Score generated by the CNN2D model is 35%.Conclusion from this research is that EAC-GAN model works better in fraud transaction in debit cards surpassing CNN2D model.