Integrating Blockchain and AI for Secure and Scalable Financial Transaction Monitoring

Abstract

The global financial system processes trillions of dollars in transactions daily, creating an urgent need for monitoring systems that are secure, scalable, privacy-preserving, and capable of detecting sophisticated fraud in real time. Despite a decade of development, existing approaches are fundamentally fragmented: rule-based anti-money laundering (AML) systems generate excessive false positives, centralized machine learning models introduce data privacy risks, and legacy database architectures create single points of failure. This paper proposes and evaluates an integrated framework that combines permissioned blockchain infrastructure with federated machine learning to address these limitations holistically. The proposed architecture anchors transaction records on a Hyperledger Fabric blockchain, reducing reliance on centralized points of failure and ensuring tamper-evident auditability. Concurrently, federated gradient-boosted and graph neural network (GNN) models detect anomalous behavior without requiring raw transaction data to leave institutional boundaries. Experimental results on the PaySim and IEEE-CIS datasets show that the federated GNN component achieves an AUC-ROC of 0.961 and a recall of 93.4% on fraudulent transactions, outperforming isolated institutional models and slightly exceeding the centralized LightGBM baseline. Blockchain benchmarking shows 340 ms average commit latency at over 2,000 TPS, and smart contract rules reduce manual AML alert review by an estimated 41% in simulation. Together, these results demonstrate that integrating blockchain-based auditability with privacy-preserving federated learning enables a unified, scalable, and effective architecture for next-generation financial transaction monitoring.