Check Fraud Network Detection Using Graph Neural Networks (GNNs)

(1) Partner Activities: Nasdaq Verafin is a global leader in financial crime management, providing advanced fraud detection and anti-money laundering solutions to financial institutions. The company processes millions of transactions daily, including check-based payments, to identify suspicious patterns and prevent financial crimes.
(2) Challenges: A critical challenge in check fraud detection is identifying sophisticated fraud rings that operate across multiple accounts and institutions. Traditional rule-based systems analyze transactions in isolation, missing network-level patterns where fraudsters use stolen or counterfeit checks systematically. Single-transaction analysis results in high false positive rates and fails to detect coordinated fraud schemes involving altered payees, check washing, and organized fraud rings.
(3) Anticipated Benefits: This GNN-based approach will provide Nasdaq Verafin with a powerful network-aware fraud detection capability, enabling identification of fraud rings and coordinated attacks that conventional methods miss. By analyzing checks within their relational contextincorporating image similarity, account relationships, and transactional patternsthe system will significantly reduce false positives while improving detection of organized fraud schemes. This innovation will enhance Verafin’s competitive advantage, strengthen its fraud prevention platform, and provide clients with more sophisticated protection against emerging check fraud tactics, ultimately reducing financial losses across the banking industry.

Faculty Supervisor:

Terrence Tricco

Student:

Partner:

NASDAQ Canada Inc

Discipline:

Computer science

Sector:

Finance and Insurance; Information and cultural industries

University:

Memorial University of Newfoundland

Program: