Point of Compromise Detection with Unsupervised Learning

Abstract

With the increase of credit cards usages, credit card frauds have also increased over time. Criminals have developed various methods to steal credit card data from users. In this study, a novel method is proposed to reduce credit card fraud by identifying credit card copying points(point of compromise) where credit card data is stolen by criminals. The proposed method extracts new feature space with an autoencoder. Then, K-means clustering is applied to cluster fraudulent transactions with extracted feature space in order to achieve grouping up similar frauds. Initial results show that, the proposed model has been able to detect 5 points of compromise from 18 points of compromise that have been detected by the banks based on information only on card transactions.