Javascript must be enabled for the correct page display

Hiding Money Laundering with an Intelligent Multi-Agent System Simulation

Keulen, Ingeborg van (2021) Hiding Money Laundering with an Intelligent Multi-Agent System Simulation. Master's Thesis / Essay, Artificial Intelligence.


Download (1MB) | Preview
[img] Text
Restricted to Registered users only

Download (135kB)


Detecting money laundering has become more complex due to the growing convergence and decentralization of financial services, as well as innovative payment methods. Considering that financial data is sensitive, there is a scarcity of labeled data which is often required for training AI-based models. A learning system generating smart fake money laundering (ML) transactions would make a good source of fake labeled data. The aim of this research is to investigate whether a dynamic MAS, simulating a transaction network, can learn to hide ML from static anti-money laundering methods. We compared three different architectures for acting as ML agents: a Support Vector Machine (SVM), a Double Deep Q-Network (DDQN), and a Bootstrapped DDQN (BDDQN). Our results show that the SVM model learns the most optimal solution in a small number of episodes when the most optimal solution involves repeating a single action multiple times. If the solution would be more complicated, then it would not learn a solution without getting caught. The DDQN model would learn solutions in which the agent gets caught, and it often learned a suboptimal policy. The BDDQN model performed the best overall, learning a (near to) optimal solution in most cases presented, as well as a solution in which the agent is not getting caught by the detection method. Our results show that the BDDQN model is sensitive to exploding gradients and can benefit from transfer learning, but further research is needed in this area.

Item Type: Thesis (Master's Thesis / Essay)
Supervisor name: Gattinger, B.R.M.
Degree programme: Artificial Intelligence
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 13 Aug 2021 05:43
Last Modified: 13 Aug 2021 05:43

Actions (login required)

View Item View Item