Credit Card Fraud Detection using autoencoder based clustering

Mohamad Zamini, Gholamali Montazer

Abstract: By increasing growth of e-commerce, which has coupled with the increase in online payments, fraud detection has become an important issue for banks. Fraud in financial transactions can cause heavy damages and endanger their reputation among their customers. Thus, focusing on a variety of fraud detection methods, as well as new ways to tackle and preventing them, is becoming increasingly important. In this paper, we have proposed an unsupervised fraud detection method using autoencoder based clustering. An autoencoder with three hidden layer and a k-means clustering has been used and tested on 284807 transactions from European banks. Based on the results, the accuracy of this method was 98.9%, as well as 81% TPR which outperforms in comparison with others.