How Nvidia Solutions are Powering AI Fraud Detection

With research finding that credit card transaction fraud could cost financial institutions $43 billion by 2026, graphics processing and artificial intelligence leader Nvidia has released a new fraud detection system to help detect fraudulent transactions.

The system, announced at the Money20/20 conference in Las Vegas, uses accelerated data processing to identify patterns in transaction data that traditional fraud detection systems might miss. Powered by the Nvidia AI platform on Amazon Web Services, it marks a shift from conventional fraud detection methods towards more sophisticated machine learning approaches.

“Financial institutions need cost- and energy-efficient computing power to handle the massive volumes of data required to fight fraud effectively,” says Pahal Patangia, global developer relations lead for consumer fintech at Nvidia.

Technical innovation aims to reduce false positives

The system combines gradient-boosted decision trees – algorithms that make predictions by combining multiple smaller models – with graph neural networks, which analyse relationships between data points. This combination aims to reduce false positives in fraud detection, a persistent challenge for financial institutions that can lead to customer frustration and lost revenue.

Nvidia’'s RAPIDS Accelerator for Apache Spark processes the data, while the Morpheus framework examines incoming transactions and flags suspicious activity. The Triton Inference Server, a deployment platform for AI models, manages the real-time processing.

The integration of these components into a single software package represents a departure from the fragmented approach many financial institutions currently use, with financial organisations using machine learning tools reporting up to 40% improvement in fraud detection accuracy.

Industry adoption of fraud detection system

Some of the world's leading financial institutions have been using AI to build proprietary solutions that mitigate fraud and enhance customer protection. Payments and credit card company American Express has integrated Nvidia’s AI platform into its fraud detection system, which monitors global transactions in real time. Nvidia says the company began using AI for fraud detection in 2010 and has since expanded its capabilities to generate fraud decisions within milliseconds.

Bunq, the European digital bank, also reports its AI-powered transaction monitoring system achieved faster model training speeds after implementing Nvidia’s accelerated computing technology. The bank uses generative AI (Gen AI) and large language models – AI systems trained on vast amounts of text data – to detect fraud and money laundering.

Meanwhile the Bank of New York Mellon, a global investments company, became the first major bank to deploy Nvidia’s DGX SuperPOD system with DGX H100 processors in March 2024, aiming to enhance its fraud detection capabilities. The DGX SuperPOD is a cluster of high-performance computing systems designed for AI workloads.

Market impact

The release comes as North American financial institutions report continued increases in online and mobile fraud losses. Nvidia’s system’s ability to process large-scale datasets and deliver real-time AI performance could help address this trend.

While the system currently focuses on credit card transaction fraud, Nvidia states it could be adapted for other types of financial crime, including new account fraud and money laundering. This versatility could appeal to financial institutions looking to consolidate their fraud detection systems.

The platform enables financial institutions to transition their fraud detection workflows from traditional computing infrastructure to accelerated computing using the Nvidia AI Enterprise software platform and Nvidia GPU instances—specialised processors designed for parallel processing tasks.

"As AI models expand in size, intricacy and diversity, it’s more important than ever for organizations across industries — including financial services — to harness cost- and energy-efficient computing power” says Patangia.

“Traditional data science pipelines lack the necessary compute acceleration to handle the massive volumes of data required to effectively fight fraud amid rapidly growing losses across the industry," he adds. "Leveraging RAPIDS Accelerator for Apache Spark could help payment companies reduce data processing times and save on their data processing costs.”

Make sure you check out the latest edition of FinTech Magazine and also sign up to our global conference series - FinTech LIVE 2024

FinTech Magazine is a BizClik brand