I-Corps: Mobile Application for Preventing Credit/Debit Card Fraud in Real Time

North Carolina Agricultural & Technical State University, Greensboro NC

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project is to provide the consumers of credit/debit cards with tools for preventing fraud in real-time at the point of contact, that is, before it posts to the account. Credit/debit card transactions involve five key players, namely: payment network provider, issuing bank (issuer), acquiring bank, merchant and cardholder (consumer). The payment network providers are the target customers of this project. They facilitate transactions and provide services to issuers and acquirers. The project proposes a payment defense strategy with the potential to save customers' money from fraud losses by placing the consumer between the merchant and the issuer, thereby adding another layer of security to the chain of communication. The project shifts the responsibility of fraud prevention from banks to the consumers by notifying consumers of impending fraud and providing the requisite techniques to stop the fraud, thereby reducing the time to launch a successful attack. Solutions on the market are reactionary in nature and rely on consumers to detect and report fraud to issuers. Issuers then perform forensic analysis to verify the legitimacy of the claim. These solutions aim to prevent future attacks whereas the proposed solution aims to prevent the attacks at hand. This I-Corps project presents a real-time, proactive solution that tailors machine learning (ML) with human intelligence to prevent the fraud from occurring. The project implements an adversarial ML technique to simulate card usage behaviors to realistically predict changing behaviors and distinguish them from potential fraud. The ML module is hosted on the back-end server and communicates with payment networks via APIs (application programming interfaces) to obtain real-time transactional data from merchants. This same data is sent to issuers to verify the availability of funds. The acquired data is scored to determine whether it is fraudulent. Scores are compared to a set of threshold values, beneath which consumers are notified for manual approval/declining of the transaction by a single click on their mobile phone. Above this threshold, the transaction is automatically approved or disapproved, signifying a legitimate or fraudulent transaction, respectively. In the case where consumers make manual inputs to approve or decline transactions, transfer learning is used to learn new behaviors to improve the performance of the existing model. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.