Money Mule Detection

Industry Partner:

Swissquote

Description:

Fraud is a significant problem for users and financial institutions, resulting in high losses. To combat fraud, most banks today leverage machine learning (ML) models. However, the industry’s existing techniques are unable to detect fraud sufficiently. The goal of our research is to develop improved ML solutions with stronger fraud detection capabilities, thereby reducing losses for users and financial institutions.

The particular focus of this project is on detection of money mules which have become an essential part of organized crime. Criminal organizations use attack vectors such as phishing campaigns to take over financial accounts. Using compromised accounts, financial assets are sent to money mules, that is, individuals who have been hired to assist in the transfer and laundering of stolen funds. After several such payments hops, the stolen funds typically find their way to cryptocurrency exchanges, offshore accounts and similar locations.

The goal of our research is to detect such malicious financial activities. While recent developments in machine learning provide a good starting point for our research, there are also significant challenges, such the scarcity of labeled training data, short active lifetime of money mule accounts, and constantly changing tactics of criminal organizations. In this research, we tackle the following questions: Professional criminal organizations are likely to adapt their fraud patterns as new detection techniques are introduced – what is the best fraud strategy that remains
for fraudsters when an improved fraud detection ML system is implemented? How exactly do the strategies of fraudsters change over time when a fraud detection system is put in place? Can money mule detection systems be adjusted to increase the economic costs for fraudsters? Are there ways to make certain types of fraud economically unattractive?