Journal of Theoretical & Computational Science
Open Access

Abstract

Anti-Money Laundering (AML) Detection Platform Leveraging Federated Learning with NVIDIA Federated Learning Application Runtime Environment (FLARE)

Venkatesh Upadrista*, Nitin Bhargava and Ram Gopal

Money laundering remains a significant challenge to the global financial system, employing complex and evolving methods that outpace current regulations. Despite the implementation of Anti-Money Laundering (AML) compliance measures such as Know Your Customer (KYC) and Customer Due Diligence (CDD), sophisticated laundering schemes continue to exploit gaps in existing systems. Traditional rule-based monitoring systems often result in high false positive rates, leading to inefficiencies and increased operational costs. While machine learning has been employed to enhance anomaly detection, issues such as imbalanced datasets, frequent false alarms and limited adaptability to new money laundering tactics still persist. To effectively combat money laundering, there is a need for more advanced, collaborative solutions that can adapt to emerging threats.

This research introduces an advanced AML detection platform utilizing the NVIDIA Federated Learning Application Runtime Environment (NVIDIA FLARE), integrating Graph Neural Networks (GNN) and eXtreme Gradient Boosting (XGBoost) machine learning models. The platform is designed to foster collaboration among financial institutions by encouraging them to share insights on AML threats and experiences through a common model without sharing raw data. This approach allows banks to collectively improve a shared model, enabling it to learn from the money laundering incidents faced by one institution so that similar incidents can be prevented at other banks, thereby mitigating overall risk and reducing financial losses. Federated learning enables the creation of such a platform without centralizing data, significantly enhancing detection capabilities through greater data diversity, reduced biases and increased adaptability to emerging money laundering patterns.

The developed platform achieved a detection accuracy of 96.22%, demonstrating the effectiveness of federated learning in enhancing AML detection while ensuring compliance with data privacy regulations. By promoting inter-bank learning rather than isolated operations, the platform incorporates the collective knowledge of money laundering attacks into the shared model, preventing similar impacts across institutions. This collaborative effort allows banks to reduce the overall impact of money laundering incidents and enhance the resilience of the financial system.

Published Date: 2024-11-15; Received Date: 2024-10-15