Abstract
The present disclosure relates to a system and method for predicting merchant-level fraud spikes in payment transaction networks using Temporal Graph Neural Networks (TGNN). The present disclosure suggests receiving historical payment transaction data associated with a plurality of transactions. The present disclosure also suggests generating, from the historical payment transaction data, a plurality of heterogeneous transaction graphs representing transaction relationships across one or more time intervals. Node feature representations and edge feature representations are generated for the heterogeneous transaction graphs based on the transaction data. Subsequently, the present disclosure suggests determining node-specific temporal context information using transaction timestamps associated with edges connected to respective nodes. The heterogeneous transaction graphs, node feature representations, edge feature representations, and temporal context information are processed using a temporal graph neural network to learn temporal and structural fraud patterns. Further, based on the learned representations, the present disclosure suggests predicting one or more merchants likely to experience a fraud spike during a future time period and generating an output indicative of the predicted fraud spike for proactive fraud mitigation.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
U, Gayathri and Panigrahi, Sitaram, "RISK MANAGER PREDICTIVE FRAUD FUTURES", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/11381