Data Driven Approach to Anti Money Laundering

Project Title

Data Driven Approach to Anti Money Laundering

Project Area of Specialization Artificial IntelligenceProject Summary

We propose a Deep Learning approach to detect money laundering by mapping structural and behavioral patterns of money laundering industries. We will use modern deep learning methods for forensic analysis of financial data. 

We will provide exploratory visualizations of the data and will try to minimize the false alerts which the traditional rule-based systems generate at a very high frequency. This makes it very difficult for human agents to check out all those transactions and hence potentially fraudulent or harmful transactions get missed. Our major goal is to improve accuracy and minimize false positives through advanced machine learning and deep learning algorithms. A System Interface will also be created to provide AML Detection on a Dump of Transactional Data.

Project Objectives

- Research on Digital Financial Transactions. 

- Research on using detecting Fraud and Money Laundering using Machine Learning Models.

- Research on using detecting Fraud and Money Laundering using Machine Learning Models.

- Develop a System Interface that will utilize Machine Learning and Deep Learning solutions we developed to detect Money Laundering Patterns in Financial Transactions.

Project Implementation Method

Money laundering is done in a variety of fashions and criminals are continuously exploring new ways for it. Therefore, this problem is very genuine and dynamic in this era. We have proposed a system that struggles with this challenge in real-time, and at the same time breaks the limits of concept drift with a continuous learning mechanism. 

The project aims to create an intelligent system that can detect anomalies in runtime transactions that can lead to potential money laundering. Money laundering is done by breaking a large transaction into smaller chunks and performing it in chunks or with several slave accounts. This money is ultimately transferred to a certain foreign account or a local account. Another way is directly transferring a very big amount to a certain account. Our system will learn different patterns formed by fraudulent transactions and will alert any potentially fraudulent transactions in a real-time transaction environment. 

We will be simulating Synthetic Data for Financial transactions with AMLSim - The AMLSim simulator generates a series of banking transaction data together with a set of known money laundering patterns, that can be used for testing machine learning models and graph algorithms.

We will then perform Exploratory Analysis and Data visualization to help better understand the Data and how normal transactions and Fraudulent Transactions are carried out. 

We will then try to use algorithms such as GCN [Kipf and Welling, 2017] and FastGCN [Chen et al., 2018] to train a model for Fraudulent transactions Detection.

In the end, we will report our findings and Develop an interface that can be used to detect fraudulent transactions.

Benefits of the Project

As technology is growing without leaps and bounds, new and advanced modes of money transfer are being developed. With the increase in such platforms, criminals, terrorists, and other money launderers are exploiting it to explore new ways for money laundering. The software programs which is currently being used in banks to find fraudulent or suspicious activities rely on rule-based programming (if-then-else) to detect and alert criminal activity. For example, if a transaction exceeding a seven-digit figure is made between foreign capitals, it gets detected and sent to security experts for review. However, these software programs are of little use for stopping criminal organizations and similar groups, which disperse small payments across multiple continents. These smart transactions do not raise any alerts in Anti Money Laundering systems.

Artificial Intelligence or more precisely Deep learning, like in many other fields, also has been shown to be useful in Financial Institutions. Modern methods can help actively monitor suspicious and potentially fraudulent transactions. Institutions updating their system with ML/DL can reduce exhaustive human work to perform routine tasks, reduce the total time taken for triaging red flags, and allow compliance personnel to focus on more valuable and complex activities. Human involvement in the AML process is crucial; in fact, human/AI hybrid models, semi-supervised learning, and processes should enable AML transaction monitoring to take a step forward in both the efficiency and effectiveness of alert operations teams. As fintech is progressing, regulations for the banks and transactions are rapidly changing. AI-driven solutions for AML will also reduce software cost as old methods required version updates to add new rules, modern solutions would be able to use human annotations on new transactions to make the system learn new patterns and changes.

Technical Details of Final Deliverable

Our Project is research on a Data-Driven Approach to Anti Money Laundering, We have gathered Data Sources, Paysim DataSet, and AMLSIM. AMLSIM is a simulation tool for Financial Transactions. This Simulation tool is very computationally Demanding and needs a powerful System and GPU to generate a useable amount of Data. 

We will be using the following Machine Learning Models on the two datasets:

- Isolation Forest Outlier Detection

- Gaussian Naive Bayes Classifier

- Decision Tree Classifier

- Support Vector Machine Classifier

- Random Forest Classifier

We will also be working on a Deep Learning-Based Approach using Graph Convolutional Networks. This Approach is extremely GPU intensive and we need resources for efficient research on this domain. The papers we will be following are very recent for implementation of this Deep Learning Technique.
- GCN [Kipf and Welling, 2017]

- FastGCN [Chen et al., 2018] 

We will then build a system that will be able to detect Money Laundering Behaviour in Financial Transactions.

Final Deliverable of the Project Software SystemCore Industry FinanceOther IndustriesCore Technology Artificial Intelligence(AI)Other Technologies Big DataSustainable Development Goals Industry, Innovation and Infrastructure, Reduced Inequality, Peace and Justice Strong InstitutionsRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	70000
GPU For deep learning	Equipment	1	70000	70000