Methodology
Full Financial (Credit and Debit Card) Fraud Detection Using Machine Learning Classifiers
- Identifier
AYENI-2026-0013 - Version1
- Published2026-06-24
- AccessTier 1: public
Description #
The methodology paper behind the transaction fraud detection half of this initiative: a comparative study of four supervised classifiers, Decision Tree, K Nearest Neighbors, Logistic Regression, and Random Forest, on a 6.36 million record synthetic mobile money transactions dataset at a fraud prevalence of 0.0536 percent. Decision Tree and Random Forest tie for the best ROC AUC at 0.84, and Random Forest achieves the strongest balance of precision and recall for the fraud class, an F1 score of 0.74, with the best calibrated probability estimates of the four. Fraudulent transactions occur exclusively in the transfer and cash out categories, never in payment, debit, or cash in, which the companion feature engineering specification carries forward as a structural signal.
Files #
| Path | Size | SHA-256 |
|---|---|---|
files/full-financial-credit-debit-card-fraud-detection.pdf |
289,954 bytes | cad35307e29c9689b76ab8a889e1b6d275792ce1f01273e71239f2e50d92cf64 |
Each file's SHA-256 is listed above. To confirm a download is unmodified: shasum -a 256 filename
Licence: CC BY 4.0