- Published
- Research
- AI/ML
- Team
- 2024
Credit card fraud: 4 models × 8 balancing techniques
Which balancing technique helps when only 0.172% of 284,807 transactions are fraud? Published at ICCSCI 2024, presented orally.
How it works
- 01284,807 card transactions, 0.172% fraud
- 024 algorithms × 8 balancing techniques
- 03Judged by F1 on the fraud class, not accuracy
- 04Best: XGBoost with random oversampling, F1 92.4%
Best fraud-class F1XGBoost with random oversampling; 284,807 transactions, 0.172% fraud.
Best fraud-class F1
XGBoost with random oversampling; 284,807 transactions, 0.172% fraud.
The problem
Fraud is rare: 0.172% of the transactions in the dataset. A model that never flags anything is still over 99.8% accurate, so accuracy says nothing. The real question is which algorithm and balancing technique catch fraud.
My part
Co-author (2nd of 4) of the published paper, and the one who presented it.
Second of four authors, and the oral presenter at ICCSCI 2024.
Limitations
- One public dataset; results may not carry over to another bank's data.