Ashirth Barthur, Security Scientist, H2O, at MLconf Seattle 2017

H2
O.ai
Machine Intelligence
Anti-Money Laundering Solution

H2
O.ai
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.

H2
O.ai
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.

H2
O.ai
Solutions do exist. Right?
1. Yes.
2. But are limited
3. Limited due to current rule-based, stateless approach

H2
O.ai
So what do we do?
1. Up the game
2. Make the detecting systems smarter

H2
O.ai
One Solution is to use Machine Learning
- Artificial Intelligence

H2
O.ai
AML Solution Evolution
Rule-based
Model
Feature-based
Model
Pure Data Driven
Model

H2
O.ai
Rule-based Model
Alerts from
rule-based
system
Analytical Inputs:
1. LexisNexis
2. Accounts Database
3. Transaction Database
4. Card Database
Alert Decision:
Suspicious
Alert Decision:
Not Suspicious

H2
O.ai
Rule-based Model: Limitations
1. Manual analysis by an investigator
2. Dispersed datasets
3. Subjective and inconsistent
4. Time consuming
5. High false positive rate

H2
O.ai
Features - (Used in Feature-based Model)
1. Features are meta data (Extracted from the data)
2. They help algorithms capture information from the data.
3. Feature engineering is a form of language translation: Between raw data
and the algorithm.

H2
O.ai
Source of Features
1. Transactions - or payments databases
2. Account Information - customer focused database
3. Alerts - AML alerts database.

H2
O.ai
Features - Example
average balance of last 7 days
7 Days

H2
O.ai
Features: Advantages
1. Designed Features Highlight Transactional Behaviour
2. Features Continuously Track Transactional Behaviour of an account
3. Rules Variables can only Identify Threshold Changes

H2
O.ai
Feature-based Model
Alerts from rule-based
system
Alert Decision:
Not Suspicious
H2O Machine
Learning Algorithm
Alert Decision:
Suspicious
Analytical Inputs:
1. Transaction Data
2. Account Data
3. Card Data etc.

H2
O.ai
Feature-based Model: Advantages
1. Uses AI - artificial intelligence
2. AI with features uses a consistent and objective approach
3. Quick classification
4. Low false positive rate - tweaked based on risk appetite.

H2
O.ai
Feature-based Model Workflow
Alerts from rule-based
system
Alert Decision:
Not Suspicious
H2O Machine Learning
Algorithm
Alert Decision:
Suspicious
Analytical Inputs:
1. Transaction Data
2. Account Data
3. Card Data etc.
AML Analyst
Alert decision sampling by the analyst
Algorithm tuning by analyst after alert
decision sampling

H2
O.ai
Pure Data-driven Model
Not a suspicious
transaction
H2O Machine
Learning - Deep
Learning Algorithm
Suspicious
Transaction
Transaction Data
Alert Data
Card Data
Account Data

H2
O.ai
Pure Data-driven Model: Advantages
1. The algorithm understands malicious behaviour through data
2. Algorithm is smart to work without features - metadata
3. Does not need alerts for training
4. Helps in identifying any kind of anomalous behaviour
5. Deeper insights about customer

H2
O.ai
Thank You
Questions?

Ashirth Barthur, Security Scientist, H2O, at MLconf Seattle 2017

More Related Content

What's hot

Viewers also liked

Similar to Ashirth Barthur, Security Scientist, H2O, at MLconf Seattle 2017

More from MLconf

Recently uploaded

Ashirth Barthur, Security Scientist, H2O, at MLconf Seattle 2017