Real-time Fraud Detection with Graph Neural Network on DGL

This solution provides a real-time anti-fraud system architecture based on deep learning graph neural network. The solution can be used in various scenarios, such as e-commerce websites, live video, games, and social applications. The solution detects abnormal business behaviors, such as fraud transactions, data crawling, and false loan applications.

The solution architecture is built with Amazon Neptune, Amazon SageMaker, Amazon Glue, etc. The core of AI is implemented using the Deep Graph Library (DGL) library, which is an open source package built for easy implementation of graph neural network model family, on top of deep learning frameworks. The deep learning technology behind the solution is backed by Amazon Machine Learning Solutions Lab, which aims to bring you the most cutting-edge technology.

The solution comes with an end-to-end demo which simulates the financial fraud detection. The demo uses the tabular data of the IEEE-CIS open dataset to construct a heterogeneous graph and trains the graph neural network (GNN) model for the real-time fraudulent transaction detection. From the demo dashboard, you can see the hundreds of thousands of online transactions are generated and the suspicious transactions are detected.