07
DataTour 2026 — Fraud detection
Team data hackathon: detecting fraudulent transactions in an imbalanced dataset, where fraud is rare and accuracy means nothing.
PR-AUC ≈ 0.35 (team)
Stack
Python
What happened?
- 01
The question
How do you spot fraud when it is rare among normal transactions?
- 02
The result
A score around 0.35, like most teams: the problem was hard for everyone. Not glorious, but a real lesson about imbalanced data.