FINS Fraud Analysis Application - Source

This asset is a component of the MuleSoft Accelerator for Financial Services, which accelerates the implementation of essential integration use cases.

The solution boosts time-to-value in your customer 360 initiatives with pre-built APIs, templates, and reference architecture for financial services. Our foundational assets are expertly designed to jumpstart your top integration and business initiatives so that IT teams can spend more time innovating and less time building from scratch.

This asset is a dockerized Python application to analyze a transaction and predict whether it is potentially fraudulent.

Asset implementation

Trees are good at handling simpler problems with relatively linear divisions between data. While a complex model, such as a neural network, might have worked, tree models are small and require very little hardware to run once built making them an ideal choice for a small microservice. XGBoost also showed promise in other research on this problem, so it made sense to approach from a similar direction as others.

XGBoost is an advanced tree algorithm that uses gradient descent to construct weak learners in order to optimize the accuracy of our overall model. This model shows quite a bit of promise with a 99+% AUPRC score.

The code is implemented in Python behind a Flask API served by WSGI. All of the code is isolated within a Docker container for easy deployment and integration.

Deploying and configuring the application

The application can be deployed as a Docker container as follows:

• Run the docker-build.sh script to build and tag the image from the supplied Dockerfile.
• Run docker-run.sh to deploy the container and setup network ports.
• Use curl http://localhost:8000/healthcheck to get health status of the application.