Anti-Money Laundering by Group-Aware Deep Graph Learning

Abstract

Money laundering (ML) is a serious challenge that supports organized and transnational crime, with far-reaching impacts on a country’s economy, governance, and social welfare. Financial institutions, which manage the flow of money, Financial institutions have become essential partners in the global battle against money laundering. While traditional AML systems typically assess transactions one by one, they often fail to detects the large patterns that emerges when criminal groups operate in coordination. Today, money laundering is often orchestrated by organized networks rather than individuals acting alone. Recognizing this shift, a deep graph learning model now makes it possible to detect collaborative money laundering by zooming in on group dynamics and shared behaviors. Our approach models users and their transactions as interconnected nodes within a graph, using a community-based encoder to capture group dynamics and behavioral patterns. Additionally, we apply a local feature enhancement method to identify and group similar transactional behaviors, helping to uncover hidden laundering networks. This Experiments are using an actual data from a leading international bank card network revealed that this approach delivers notably higher accuracy in detecting suspicious activity. It delivers superior results compared to current anti-money laundering methods, consistently identifying suspicious activity more effectively in both live monitoring and scheduled data analysis. These findings underscore how using graph-based techniques to capture group-level patterns can make AML systems far more effective at spotting complex, coordinated financial crimes.

Introduction

1. Money Laundering Overview

Money laundering involves disguising illegally obtained money to make it appear legitimate. It supports organized crime and is estimated to account for about 2.7% of global GDP. Traditional anti-money laundering (AML) methods are rule-based and labor-intensive. New technologies like machine learning (ML) and deep learning are now being adopted to automate and improve detection—but they often struggle to detect coordinated laundering activities by organized groups.

2. Proposed Approach: Group-Aware Graph Neural Network (GAGNN)

To overcome limitations of traditional AML models, researchers propose a Group-Aware Graph Neural Network (GAGNN) that:

• Models accounts as nodes and transactions as edges in a graph.

• Captures both individual behavior and group dynamics.

• Uses Graph Attention Networks (GAT) and Extended Markov Random Fields (eMRF) to identify suspicious communities.

• Classifies suspicious activity at three levels:

  1. Individual accounts

  2. Transactions

  3. Groups of users (super-nodes)

This multi-level structure enables better detection of organized money laundering schemes. It is scalable, trained offline, and suitable for real-time deployment.

3. Literature Review – Key Techniques and Contributions

a. Time-Frequency Analysis (El Ammari et al., 2021)

Uses Short-Time Fourier Transform (STFT) to detect laundering patterns over time. Captures dynamic, repetitive financial behavior with high accuracy.

b. Deep Learning Models (Ghosh et al., 2023)

Applies models like RNNs, LSTMs, CNNs, and GNNs for complex pattern recognition. Highlights the importance of explainable AI (XAI) tools (e.g., LIME, SHAP) to improve transparency.

c. Rule-Based Detection (Butgereit, 2021)

Uses rule-based logic (e.g., 12 Red Flags) to detect funnel accounts in mobile finance, effective where labeled data is scarce.

d. Visual Analytics for Crypto AML (Wang et al., 2022)

Introduces a visual framework using transaction flows and clustering to detect laundering in cryptocurrency exchanges—helpful for non-technical investigators.

e. Hybrid Model (Koo et al., 2024)

Combines an autoencoder for anomaly detection with a risk-based scoring system, balancing unsupervised learning with expert-driven rules.

f. Explainable AI via xGEMs (Dou, 2020)

Provides interpretable counterfactuals using generative models to explain decisions made by black-box ML models, enhancing trust in AI systems.

g. Graph-Based Detection with Sparse Data (Liu et al., 2020)

Presents a scalable Graph Convolutional Network (GCN) that learns from limited labeled data using sampling and embedding. Effective for large, sparse financial graphs.

4. Key Innovations of GAGNN

• Graph-based transaction modeling

• Group-level analysis for detecting coordinated crime

• Community-aware attention and inference

• Multi-layer classification (users, transactions, groups)

• Scalable and real-time capable system

Conclusion

The growing complexity of financial crime demands more intelligent, adaptable, and interpretable solutions—deep graph learning offers a promising path forward in this effort. By modeling relationships between entities and capturing group behaviors within financial networks, graph-based approaches, This approach significantly surpasses the limitations of conventional anti-money laundering methods, offering deeper insights and more adaptive solutions. This survey has highlighted a range of innovative methods, from group-aware neural networks and semi-supervised frameworks to autoencoder-based anomaly detection and explainable AI tools. Each contributes uniquely to detecting both individual and collaborative money laundering patterns in real-world settings. Despite notable progress, challenges remain. Limited access to reliable data, concerns about how models make decisions, and challenges in scaling them up are still major roadblocks to broader adoption.. However, recent advancements in hybrid learning models, interpretability techniques, and scalable graph architectures suggest a promising future for research and practical deployment. Going forward, collaboration between financial institutions, regulatory bodies, and the research community will be essential to build AML systems that are not only accurate but also trustworthy and operationally viable. Ultimately, deep graph learning is not just a technical innovation—it’s a foundational shift in how we understand and combat financial crime in the digital age.

References

[1] Liu, H., Zuo, Y., Zhu, X., Yin, H., & Zhang, M. (2021). Anti-money laundering by group-aware deep graph learning. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining (pp. 1113–1123). ACM. [2] B. Unger, \"Money laundering regulation: From al Capone to al qaeda,\" in Research Handbook on Money Laundering Anonymous 2013, . [3] M. El Ammari, H. Rachidi, and Y. Benslimane, \"A time-frequency based suspicious activity detection for anti-money laundering,\" in Proc. 2021 4th International Conference on Advanced Systems and Emergent Technologies (IC_ASET), Hammamet, Tunisia, 2021, pp. 177–182. doi: 10.1109/IC_ASET52486.2021.9691095. [4] A. Ghosh, M. V. R. K. R. R. Atrey, and A. B. Benerjee, \"Deep learning and explainable artificial intelligence techniques applied for detecting money laundering: A critical review,\" Computers & Security, vol. 130, 2023, Art. no. 103138, doi: 10.1016/j.cose.2023.103138. [5] L. Butgereit, \"Anti Money Laundering: Rule-Based Methods to Identify Funnel Accounts,\" in Proc. 2021 Conf. on Information Communications Technology and Society (ICTAS), Port Elizabeth, South Africa, 2021, pp. 20–26. doi: 10.1109/ICTAS50802.2021.9394990. [6] Wang, H., Wang, X., Yang, D., & Wu, Y. (2022). “Visual analysis of money laundering in cryptocurrency exchange.” IEEE Transactions on Visualization and Computer Graphics, 28(1), 98–108. [7] K. Koo, M. Park, and B. Yoon, \"A suspicious financial transaction detection model using autoencoder and risk-based approach,\" IEEE Access, vol. 12, pp. 68926–68939, May 2024, doi: 10.1109/ACCESS.2024.3399824. [8] Yingtong Dou1, Zhiwei Liu1, Li Sun2, Yutong Deng2, Hao Peng3, Philip S. Yu1 ‘‘Enhancing Graph Neural Network-based Fraud Detectors against Camouflaged Fraudsters,’’. 1. IEEE, 2020. [9] Liu, A., Tong, Y., Li, Z., Deng, K., He, X., & Tong, H. (2020). “Scalable semi-supervised graph learning techniques for anti-money laundering”. arXiv preprint arXiv:2009.05344 [10] T. N. Kipf and M. Welling, ``Semi-supervised classification with graph convolutional networks,\'\' in Proc. Int. Conf. Learn. Represent. (ICLR), 2017, pp. 114. [11] R. Liu, X.-L. Qian, S. Mao, and S.-Z. Zhu, ``Research on anti money laundering based on core decision tree algorithm,\'\' in Proc. IEEE Chin. Control Decis. Conf. (CCDC), May 2011, pp. 4322 4325. [12] Z. Chen, L. Dinh Van Khoa, A. Nazir, E. N. Teoh, and E. K. Karupiah ``Exploration of the effectiveness of expectation maximization algorithm for suspicious transaction detection in anti money laundering,\'\' in Proc. IEEE Conf. Open Syst. (ICOS), Oct. 2014, pp. 145 149. [13] K. Singh and P. Best, “Anti-money laundering: Using data visualization to identify suspicious activity,” Int. J. Accounting Inf. Syst., vol. 34, no. 3, pp. 7–13, 2019. [14] N. Heidarinia, A. Harounabadi, and M. Sadeghzadeh, “An intelligent anti-money laundering method for detecting risky users in the banking systems,” Int. J. Comput. Appl., vol. 97, no. 22, pp. 35–39, Jul. 2014. [15] M.-J. Segovia-Vargas et al., “Money laundering and terrorism financing detection using neural networks and an abnormality indicator,” Expert Syst. Appl., vol. 169, 2021, Art. no. 11447

Copyright

Copyright © 2025 Shwetha A B, Shreyas H G, Siddarameshwara J, Srinidhi M, Srujan T S. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.