03-Amazon-Fraud-Detector-solution

Detect fraud activities using Amazon Fraud Detector

Build, deploy, and manage fraud detection models without previous machine learning (ML) experience.

How it works

Amazon Fraud Detector is a fully managed service enabling customers to identify potentially fraudulent activities and catch more online fraud faster.

Demonstration

In this demonstration, I use boto3 SDK to build Fraud detector utilizing Amazon Fraud Detector service. You can opt to build either on console or code.

When using Amazon Fraud Detector, there will be few terms and terminologies you need to understand. Let's imagine the transaction or claim scenario:

1. The entities (i.e., can be person, hospital, merchant) make an activities, or events (this can be transaction, claim, account registration).
2. This event will consists of data points, or variable, which represents the transaction type, amount, etc.
3. This event will result in an outcome, or labels, which either can be fraud or legit transaction.

Below image demonstrates the overall process with the terms.

Now, to detect fraud, usually there are 2 options:

• Domain users have predefine logics or rules to filter the events or transactions
• Data scientists have built ML model to score or predict each event or transaction

Amazon Fraud Detector has the capabilities to combine these 2 methods, known as Detector, to buld the ML models using AutoML and define the business logic and each logic produce a different outcome. For example:

• model score >= 800, outcome is fraud
• model score > 500 and model score < 800, outcome is investigate
• model score <= 500, outcome is legit

You can define various logics and outcomes based on your business objectives.

To conclude, Amazon Fraud Detector consists of processes below.