This version supersedes an earlier one, kept live for reference: AYENI-2026-0013 →

Description #

An expanded methodology paper behind the transaction and synthetic identity detection initiative, now validated across two structurally distinct payment environments. In a credit card transaction dataset of about 1.9 million records, Random Forest reaches 99.74 percent accuracy on a held out test set of 555,719 records. In an account transfer dataset of over 6.3 million records at a fraud prevalence of 0.0536 percent, Random Forest reaches 99.95 percent accuracy, an F1 score of 0.74 for the fraud class, and an AUC ROC of 0.84. The cross environment consistency of these results is presented as evidence that the fraud signal lives in the feature construction rather than in any single model architecture. Specifies a 20 measurement synthetic identity feature set across six behavioural families, designed for community banks, credit unions, and regional lenders operating without dedicated analytics staff.

Files #

PathSizeSHA-256
files/full-financial-credit-debit-card-fraud-detection.pdf 231,510 bytes 95f249b8df43f50b9b23c199a373374a105da42caf15652a280cd5f91a51cb0a

Each file's SHA-256 is listed above. To confirm a download is unmodified: shasum -a 256 filename

Licence: CC BY 4.0

← Back to Initiative 2