Return to page


Detecting Money Laundering Networks Using H2O Driverless AI


By Parul Pandey | minute read | March 05, 2020

Blog decorative banner image

Note: Dr. Ashrith Barthur (Principal Security Scientist, and Sandip Sharma (Director of Solution Engineering, will be speaking about solving money laundering and other real-world problems using machine learning at our upcoming webinar. You can grab a spot here 


Artificial Intelligence has evolved from being a buzz word to a reality today. It is making a positive impact on several industries, and the financial sector hasn’t been left untouched. The financial services industry is continuously innovating and advancing new technologies in the pursuit of increasing their customer base and finding new opportunities. This is happening across all segments from capital markets to commercial banking, and from consumer finance to insurance.

The use of AI in the financial industry is quickly changing the business landscape, even in traditionally conservative areas. Financial institutions today use artificial intelligence (AI) for scenarios like customer services, risk management, fraud detection, and anti-money laundering while adhering to regulatory compliance.

Some areas where AI is helping the business

AI solutions have proved to be a boon when it comes to detecting Money Laundering  and, the open source leader in AI, is empowering leading financial services companies to deliver AI solutions that help tackle such menace, effectively.

Anti-Money Laundering

Money laundering “is the concealment of the origins of illegally obtained money, typically by means of transfers involving foreign banks or legitimate businesses.” 

Impact of Money Laundering on Global Banking

Money laundering is a massive problem for the financial services sector. According to the United Nations Office on Drugs and Crime, an estimated $2 trillion is “cleaned” through the banking system each year. Fines for banks who fail to stop money laundering have increased by 500X in the last decade to more than $10 Billion per year. As a result, banks have built large teams of people and given them the time-consuming task of finding and investigating suspicious transactions, which often take the form of numerous small transfers within a complex network of players.

Traditional approaches for solving Money Laundering

Investigation teams have used rules-based systems like FICO , Fiserv , SAS  AML , Actimize , etc. to find suspicious transactions. The steps involved in the current rule-based workflow are as follows:

  • An alert is generated by the alerting system.
  • The investigator reviews it using information from different sources.
  • The alert is approved as True Positive or classified as False Positive.

However,rule-based systems have a large amount of false positives usually in the range of 75 – 99%. These rules can quickly become outdated thus producing large numbers of false positives that still need to be reviewed. Some of the drawbacks of using Rule-Based systems are:

Issues with Rule-based AML Approaches
How AI can address this issue

Anti Money Laundering (AML)  programs in capital markets and retail banking extensively deploy Rule-based Transaction Monitoring Systems , spanning areas like monetary thresholds and money laundering patterns. Bad actors can learn these rules over time and change their methods to avoid detection. AI-based behavioral modeling  and customer segmentation can be more effective in discovering transaction behaviors and identify behavioral patterns and outliers, indicating potential laundering.

AI, especially time series modeling, is particularly good at looking at a series of complex transactions and finding anomalies. Anti-money laundering using machine learning techniques can find suspicious transactions and networks of transactions. These transactions are flagged for investigation and can be scored as high, medium, or low priority so that the investigator can prioritize their efforts. The AI can also provide reason codes for the decision to flag the transaction. These reason codes tell the investigator where they might look to uncover the issues and help to streamline the investigative process. AI can also learn from the investigators as they review and clear suspicious transactions and automatically reinforce the AI model’s understanding to avoid patterns that don’t lead to laundered money.

Solving Anti-Money Laundering problem using Driverless AI
H2O Driverless AI

H2O Driverless AI  is an Award-winning Automatic Machine Learning Platform that empowers data scientists to work on projects faster and more efficiently by using automation to accomplish key machine learning tasks in just minutes or hours, not months. >’s AML solution uses Driverless AI as the modeling engine utilizing Driverless AI’s advanced feature engineering and model creation capabilities.’s Anti Money Laundering(AML) system provides the following advantages over the rule-based system:

  • Fundamentally reduces the False positives
  • It has the ability to ingest recipes, customized for money laundering.
  • The solution is strategically placed between the AML system and the investigator
  • It classifies alerts as False Positives, or True Positive using an out-of-loop ML approach.
  • A curated set of alerts are given to the investigator

The usual investigation time is reduced drastically from 45–90 days to seconds. It also reduces human inaccuracies and required person-hours. It fits rule-gaps with innovative features.

Demo with Driverless AI


Dataset’s AML system can work with pre Marked AML alert Data, Transactional banking data, and Banking KYC data. However, for this demo, we’ll be using a synthetic dataset whose distribution is very similar to that of a financial dataset. The dataset has The training data consists of both numeric and categorical columns. Some of the fields are -person’s account number, date, kind of business, typology, etc.

The target  column identifies whether an alert is suspicious enough to be sent for further investigation. This is something that we need to predict i.e., whether the generated alerts are useful or not. Our idea is to reduce the false positives in this case. Here it is important to know that the model learns the behaviour based on transactions, and not specific to any one person.

A Glimpse of the Dataset
Launching the Experiment

In case you want to refresh your knowledge about getting started with Driverless AI, feel free to take a  Test Drive . Test Drive is H2O’s Driverless AI on the AWS Cloud where you can explore all its features without having to download it. 

The data is ingested into a Driverless AI instance and treated as a Supervised ML problem.  Once the dataset has been ingested, we get into the expert settings and bring in the recipe specifically designed for identifying false positives in money laundering alerts.

Finally, we select F1 as the scorer(since we want to minimize the false negatives as much as possible). Further, we can adjust the accuracy; time and interpretability settings to suit our needs.

Driverless AI Screen with AML Recipe in Action

Keeping all the other parameters as default, we launch the experiment. The screen should appear as follows:

Driverless AI Experiment

Driverless AI lets us see the variable importance of the features used for the model building purpose, in addition to the F1 scores at each iteration. For this experiment, Driverless AI performed the following steps to find the optimal final model:

www.h2o.ai2020/03/parul-9.png www.h2o.ai2020/03/parul-9.png

AI is critical to success in the financial services industry. Driverless AI enables financial services companies to build personalized banking experiences quickly, fraud and money laundering models, improve employee productivity, and more.



Parul Pandey

Parul focuses on the intersection of, data science and community. She works as a Principal Data Scientist and is also a Kaggle Grandmaster in the Notebooks category.


Ashrith Barthur

Ashrith is the security scientist designing anomalous detection algorithms at H2O. He recently graduated from the Center of Education and Research in Information Assurance and Security (CERIAS) at Purdue University with a PhD in Information security. He is specialized in anomaly detection on networks under the guidance of Dr. William S. Cleveland. He tries to break into anything that has an operating system, sometimes into things that don’t. He has been christened as “The Only Human Network Packet Sniffer” by his advisors. When he is not working he swims and bikes long distances.


Sandip Sharma

Sandip is an entrepreneur and technology leader, who has a balance of work experience in both Financial Services Industry and Government. With +20 years of experience in business IT, Sandip thrives for developing and implementing innovative AI/ML solutions to the Whole-of-Government and Financial Services Industry on emerging digital technologies. He has Masters. Degree in Business IT – Financial Services, from Singapore Management University (SMU).