This Deep Learning Technology is a Money-Launderer’s Worst Nightmare

by Zaveria K November 28, 2022

LaundroGraph is using deep learning technology to support anti-money laundering efforts

Deep learning technology has shown to be extremely useful in addressing a wide range of academic and real-world challenges in recent years. Deep learning has been proved by researchers at Feedzai, a financial data science firm located in Portugal, for the prevention and detection of several criminal money laundering operations.

The Feedzai team developed LaundroGraph, a self-supervised model that might reduce the time-consuming process of assessing vast volumes of financial interactions for suspicious transactions or monetary exchanges, in a paper presented at the 3rd ACM International Conference on AI in Finance. Their approach is based on a graph neural network, which is an artificial neural network or ANN built to process vast volumes of data in the form of a graph. “Wanting to enhance our AML service, and after identifying significant pain points with the present AML reviewing process, we considered alternatives to solve these issues using AI,” said Mario Cardoso, a Research Data Scientist at Feedzai.

“AML is particularly difficult because of the paucity of labels, as well as the fact that the context around financial movements, namely the entities engaged with and the features of each transaction, is critical in informing judgments. With these limitations in mind, we set out to develop a machine learning solution that may assist human analysts while also facilitating AML review.”

Examining financial transactions Looking for suspicious activities may be a time-consuming and unpleasant effort for human analysts. Cardoso and his colleagues set out to substantially simplify this work by employing deep learning techniques, which are well-known for their ability to analyze vast volumes of data.

The model they developed, LaundroGraph, can encapsulate banking clients and financial activities, translating them into useful graph representations. These representations can help anti-money laundering analysts by exposing unusual money movements for certain clients without requiring them to examine whole transaction histories.

Cardoso explained, “LaundroGraph generates dense, context-aware representations of behavior that are decoupled from any specific labels.” “It accomplishes this by utilizing both structural and features information from a graph via a link prediction task between customers and transactions. We define our graph as a customer-transaction bipartite graph generated from raw financial movement data.”

Feedzai researchers put their algorithm through a series of tests to see how well it predicted suspicious transfers in a dataset of real-world transactions. They discovered that it had much greater predictive power than other baseline measures developed to aid anti-money laundering operations.

“Because it does not require labels, LaundroGraph is appropriate for a wide range of real-world financial applications that might benefit from graph-structured data,” Cardoso explained. “Our paper proposes to leverage these embeddings to provide insights that can accelerate the AML detection reviewing process, but this approach can be extended to other use cases (e.g., fraud), and the embeddings can serve a variety of purposes beyond the insights we analyze (e.g., feature enrichers).”

In the future, LaundroGraph might let financial analysts and anti-money laundering agents throughout the world evaluate massive quantities of financial transactions, allowing them to spot suspicious activity more quickly and effectively. Cardoso and his colleagues intend to expand on their methodology while also investigating its possibilities for tackling other financial challenges.

“Future directions for our research will include experimentation in additional use cases, such as fraud, as well as research into other insights/tasks that can be enabled or enhanced through the embeddings, such as using the embeddings as an informative starting point for label-scarce downstream predictions,” Cardoso added.

Share This Article

Do the sharing thingy

Source link