August 12, 2021

Anti-Money Laundering (AML) Enhanced with Machine Learning

Anti-money laundering (AML) laws seek to deter criminal activity that conceals illegally obtained money. The Bank Secrecy Act (BSA) of 1970 established that private individuals, banks, and other financial institutions had to keep exact records and reports.

This helps to identify the source, volume, and movement of money transported or transmitted in or out of the US or deposited in financial institutions. Banks were required to report cash transactions over $10,000, identify the person conducting the transactions, and maintain a paper trail of the transactions.

Per the Money Laundering Suppression Act of 1994, financial systems are required to monitor their customers’ transactions and report any suspicious or unusual financial activity.

Efforts to prevent money laundering, however, are not as effective as they could be. Most AML strategies today are still scenario-based. With further development of virtual pay and the trading market, new and complicated scenarios germinate.

It is time-consuming and labor-intensive to detect new scenarios and develop certain rules to fit them, especially for community crimes. The rule-based AML system has been proved to be vulnerable and easy to be cracked up.

Advanced machine learning and deep learning techniques have been developed for years. Now they are widely used in financial services, such as credit risk, anti-fraud, etc. With third-party companies putting more effort into business data analysis, many of them afford analyzed data(labels) as external resources to enhance the AML detection process.

Applying advanced machine learning techniques into AML could not only save manual efforts but also dig out some latent patterns which used to be hard to define. Cooperation with third-party for external data could expand the information sources to make the detection procedure more accurate.

An end-to-end AML product with improved detection accuracy and efficiency can be achieved through machine learning. Let’s break down how this technology works.

Data Enhancement

  • Existing data sources are enhanced using various techniques or third-party data.
  • Web Crawling
  • Public information contains formal and informal customer details; however, widely spread internet and non-standardized information is hard to collect.
  • Web crawling is a fast and automatic way to gather scattered data.
  • Feature Engineering
  • Extract useful features and make them structured and comparable.
  • Multi-dimensional description helps with abnormal transaction detection.
  • As the foundation of AML, Know Your Customer (KYC) identity verification uses a 360-degree customer profile with feature engineering techniques and a basis of AML domain expertise.

Know Your Customer (KYC)

  • As the foundation of AML, Know Your Customer (KYC) identity verification uses a 360-degree customer profile with feature engineering techniques and a basis of AML domain expertise.

Transaction Monitoring

  • Advanced machine learning techniques are used to make more accurate detections for each transaction.
  • Human efforts of reviewing suspicious transactions are greatly reduced.

Community Detection

  • Typical features are obtained from the community perspective.
  • Graph algorithms offer a data-driven method to recognize money laundering connections amongst the larger population.
  • Machine learning can teach computers to detect and recognize suspicious behavior and classify alerts following a risk-based approach as being a high, medium, or low-risk level.
  • Applying rules to these alert classifications can facilitate hibernation and auto closure of alerts. This allows employees to supervise machines that triage these alerts rather than manually reviewing the alerts.
  • Unsupervised: Determine the AML target with these machine learning algorithms: clustering, Gaussian Mixture Modelling (GMM), dimensional reduction
  • Supervised: Discover patterns and population with these machine learning algorithms: Long short-term memory (LSTM), Support Vector Machine (SVM), Neural Network, Transfer Learning

AML Solution Architect

How ElectrifAi Can Help

Our vast library of pre-built machine learning models and deep domain knowledge enables you to quickly act against money laundering violations. Our models can help detect potential community anomalies, money laundering transactions with predicted risk ranks, and provides a semi-automatic system that sends out alerts for potential money laundering activity.

Want to find out how your company can help contribute to more accurate anti-money laundering efforts? Contact us today.