anti money laundering solutions
TRANSCRIPT
H2O.aiMachine Intelligence
Ashrith Barthur, PhDSecurity Scientist at H2O.ai
Anti Money Laundering Solutions
H2O.aiMachine Intelligence
What is Money Laundering?
1. “Washing” ill-gotten money with legitimate money to hide
the source
2. Illegal drug sales, human trafficking, online gambling,
insider trading, etc.
H2O.aiMachine Intelligence
What is the problem with Money Laundering?
1. Illegal trade and markets grow
2. Negatively impacts the society
3. Governments lose out on taxes
4. In some countries, alternative centers of power come into
existence.
H2O.aiMachine Intelligence
Solutions do exist. Right?
1. Yes.
2. But are limited
3. Limited due to a rule-based, stateless approach
H2O.aiMachine Intelligence
So what do we do?
1. Up the game
2. Make the detecting systems smarter
H2O.aiMachine Intelligence
One Solution is to use Machine Learning - Artificial Intelligence
H2O.aiMachine Intelligence
Why Machine Learning/Artificial Intelligence?
1. Machines (programs) learn from data.
2. More probability of the occurrence of an event = more
probability of the machine learning and being able to
predict the event.
H2O.aiMachine Intelligence
More Data = Better AI. Right?
1. At a fundamental level, “Yes”
2. But the machines need great features (variables) to learn
the inherent behaviour
H2O.aiMachine Intelligence
Let us see that with an example use-case.
H2O.aiMachine Intelligence
Example use-case
Quick or excessive cash withdrawals using ATMs.
H2O.aiMachine Intelligence
ATM Cash Withdrawals
1. Used to laundering money quickly
2. Mostly done through unsuspecting account holder -
“Money Mules”
3. Money movement done by promising a small percentage
of the deposited money
4. Sometimes with fake and/or new accounts
H2O.aiMachine Intelligence
ATM Cash Withdrawals
1. Difficult to identify using rule-based systems
2. Rule based systems capture only trips
3. Below threshold laundering is missed by rule-based
systems
4. Rule based systems are fairly memoryless (stateless)
H2O.aiMachine Intelligence
Detecting Cash Withdrawals with ML/AI
1. Machine Learning/AI system do not depend on trips and
thresholds
2. They learn from numerous features that are designed by
engineers
3. The learn from feedback - money laundering experts, what
is money laundering and what is not.
H2O.aiMachine Intelligence
Feature examples
1. Age of the account
2. ATM withdrawal / All withdrawal per week - tracked for 6
months (rolling window)
3. Average volume of withdrawal per week - tracked for 6
months (rolling window)
H2O.aiMachine Intelligence
Age of the Account
1. Investigators / AML Experts are routinely wary of new
accounts
2. New accounts tend to launder money more than old
accounts
3. Our tests corroborate that.
H2O.aiMachine Intelligence
ATM Withdrawal/ All Withdrawal
1. Feature tracked for 6 months
2. The idea is that historically tracking a change or no change
helps the AI algorithm learn the behaviour
3. Actual money laundering will show spikes
H2O.aiMachine Intelligence
Average volume of withdrawal
1. Another feature tracked for 6 months
2. Changes in the volume of withdrawal is indicator of a
possible money laundering behaviour
H2O.aiMachine Intelligence
Aggregate of Features
1. No Individual feature can identify a money laundering
situation positively
2. The composition of all features yields a probability score
for a given event.
3. A high enough event probability score can confirm money
laundering.
H2O.aiMachine Intelligence
Thank YouQuestions