Closing soonData
Fraud Signal Ranking
Rank transactions by fraud likelihood for manual review.
Given anonymized transaction features, produce a ranked score used to prioritize manual fraud review. Scored on AUC-PR given heavy class imbalance.
- Solved by
- 47 people
- Highest score
- 0.783
- Submissions
- 138
- Closes
- Sep 5
Rules
- ▸No use of the transaction identifier as a feature.
- ▸Model must score the full eval set in under 60s.
Success metrics
- ▸AUC-PR above 0.65 on the held-out fraud rate.