Abstract
Computer-implemented fraud intelligence sharing systems and methods for detecting fraudulent transactions across a network while preserving member privacy are disclosed. Each network member generates transaction embeddings from confirmed fraudulent transactions, clusters the embeddings, and computes local cluster centroids representing fraud patterns. Differential privacy noise is applied locally to the centroid vectors, including dimension-wise noise and noise scaling based on cluster size, such that smaller clusters receive greater obfuscation. Only the privacy-protected centroid vectors are transmitted to a central aggregator, which computes aggregated network centroids using weighted contributions based on factors such as cluster size and confidence. The aggregated centroids are distributed to network members, enabling detection of transactions exhibiting proximity to fraud patterns observed elsewhere in the network. Proximity, confidence, and freshness signals derived from the aggregated centroids are evaluated by a real-time rule engine to generate authorization decisions, including allow, challenge, or decline outcomes, within a low-latency transaction authorization workflow.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kala, Durga S; Xu, Jun; Barcelos, Tanner Kumar; Sonowal, Ankana; and Fu, Yidong, "“System and Method for Privacy-Preserving Federated Fraud Embedding Centroid Aggregation Across Visa Network Members Without Raw Transaction Data Sharing”", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/11790