City Research Online

Machine Learning for Money Laundering Risk Detection in Online Gambling

Charitou, C (2021). Machine Learning for Money Laundering Risk Detection in Online Gambling. (Unpublished Doctoral thesis, City, University of London)


This thesis addresses the issue of money laundering in online gambling. Over the years, the online gambling industry has evolved into one of the most profitable industries on the internet. While stringent new regulations have required the industry to become more vigilant, methods used to process proceeds from illicit activities have also advanced and have become more sophisticated. This research examines the application of machine learning for the detection of high-risk money laundering cases in online gambling. This work was part of a collaboration with Kindred Group, a major gambling operator.

Money laundering as a fraud detection problem su↵ers from the binary class imbalance issue in data mining. This research focuses on investigating data and algorithmic level techniques to provide a solution to that issue. An in-depth analysis of supervised learning algorithms is carried out and a supervised learning framework is proposed to improve the detection rate of high-risk money laundering cases relative to the existing rule-based system. Results showed immediate improvement in the identification rate. Furthermore, it examines Generative Adversarial Networks (GANs) to provide a solution to the class imbalance problem by generating new synthetic data to oversample the minority class. Our GAN-based approach outperformed popular oversampling techniques when combined with supervised learning classifiers. Building on our GAN-based architecture, we then introduce a novel generative adversarial framework, based on semi-supervised learning and sparse auto-encoders, for the detection of fraud in online gambling. Experimental results show that the proposed framework outperforms mainstream discriminative techniques without the need of generating synthetic instances. We validated our system by applying it to other domains that suffer from the binary class imbalance problem.

Finally, unsupervised anomaly detection (AD) framework based on encoder-decoder long short-term memory (LSTM-ATT) networks and Gaussian estimation is examined to discover new patterns in customer behaviours that could be related to money laundering risk, something which is not possible with a supervised framework. Our AD system is evaluated with the help of Kindred’s compliance team on specific cases. The feedback received from our research partners suggested that the detected anomalies indicated risk of money laundering and that the proposed framework can be included in their existing anti-money laundering (AML) process.

Publication Type: Thesis (Doctoral)
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments: Doctoral Theses
School of Science & Technology
School of Science & Technology > Computer Science
[img] Text - Accepted Version
This document is not freely accessible until 30 November 2024 due to copyright restrictions.

To request a copy, please use the button below.

Request a copy



Downloads per month over past year

View more statistics

Actions (login required)

Admin Login Admin Login