Train a GNN for Fraud-Ring Detection at a Payments Fintech
Overview
What this challenge is about.
You receive an anonymized transaction dataset (around 120,000 merchants, around 4 million transactions over 12 months, around 2% labeled fraud) and the team's LightGBM baseline. Construct a heterogeneous graph (merchants, devices, IPs as nodes; transactions and shared-attribute edges) using PyTorch Geometric. Train a GraphSAGE classifier and compare it against the LightGBM baseline on AUC-PR, recall at 1% false-positive rate, and time-to-flag a synthetic ring you inject. Write a 2-page memo on the maintain-vs-replace decision.
The Brief
What you'll do, and what you'll demonstrate.
Quantify whether a Graph Neural Network beats the tabular baseline at identifying coordinated fraud rings in a cross-border payments network.
Earning criteria — what you'll demonstrate
- Construct heterogeneous graphs from tabular transaction data
- Train and tune GraphSAGE for node classification at scale
- Compare GNN vs. tabular baseline fairly on fraud metrics
- Reason about the operational cost of adding graphs to a fraud stack
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Data Scientist
Owning a GNN vs. tabular comparison on a real fraud problem and writing the fraud-ops memo is exactly the day-one job of a data scientist in fintech.
This challenge sharpens
- fraud-detection
- graph-neural-networks
- evaluation
Machine Learning Engineer
Constructing graphs at scale and shipping a GNN training pipeline is core MLE work for any company that has graph-shaped data.
This challenge sharpens
- pytorch-geometric
- graphsage
- graph-construction
ML Researcher
Designing a ring-aware time-to-flag metric and benchmarking GNNs against strong baselines is the kind of methodology work ML researchers do in applied research labs.
This challenge sharpens
- graph-neural-networks
- evaluation
- graphsage