Abstract
A computer-implemented system for monitoring machine learning model input feature health in a payment transaction authorization network. The system includes a distributed audit log extraction pipeline configured to ingest ML model scoring event records, filter records by model identifier and issuer processor affiliate identifier, and flatten nested JSON request and response structures into columnar format. The system includes a transaction-type partitioning module configured to split the flattened dataset into per-transaction-type subsets corresponding to different authorization pathways. The system includes a numerical feature statistics engine configured to compute distributional statistics for numerical scoring features. The system includes a population stability index computation engine configured to compare current production feature distributions against validated baseline distributions and compute per-feature PSI values with tiered alert thresholds. The system includes a categorical feature shift detector configured to compute day-over-day distributional changes for categorical transaction attributes and generate alerts when changes exceed a configurable threshold.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kala, Durga S.; Xu, Jun; Sonowal, Ankana; and Fu, Yidong, "ADAPTIVE REAL-TIME ML MONITORING, DRIFT DETECTION, CROSS-ENVIRONMENT PSI, AND HYPERSCALE AUTHORIZATION FOR MULTI-RAIL PAYMENT FRAUD PREVENTION", Technical Disclosure Commons, (August 26, 2026)
https://www.tdcommons.org/dpubs_series/11487