Fraud detection in OpenPayments data

Inspiration

Being new in the Interledger/Open Payments ecosystem one of the things that I felt it was missing while casually browsing and testing out rafiki is the lack of some sort of refund reporting method that is common in the mainstream banking portals.

I thought that since Open Payments provides us with a spec for transactions it would be interesting to prototype a ML pipeline as a draft for detecting malicious transactions.

What it does

Provides some basic functionality to test, train, evaluate and serve a ML model that can classify a transaction as fraud.

How we built it

The core of the functionality is Ludwig an AI prototyping framework based on pytorch.

Challenges we ran into

• My test environment didn't have enough data to prototype so we used mock data instead for the purpose of prototyping
• I didn't completely understand the lifecycle of a transaction in openpayments and if/how we can update a payment.
• Initially I wanted to provide some sort of UI to provide a way for users to report a potentially malicious transaction as a way to input data to the ML training process but this ended up being more complicated than I initially thought.

Accomplishments that we're proud of

The core functionality for building/testing/evaluating/serving the model is there.

What we learned

I learned a lot for Open Payments API and its usages also this was a good gateway to understand the rest of the parts of the Interledger/Open Payments ecosystem.

What's next for Fraud detection in OpenPayments data

Building a prototype with a user facing UI workflow based on rafiki.

Built With

• python
• pytorch