Deep Learning-Driven Optimization of ISO 20022 Protocol Stacks for Secure Cross-Border Messaging

Abstract

The ISO 20022 protocol stack forms a pillar of the digital financial ecosystem by enabling cross-border payment messaging between banks. Nevertheless, the presence of multiple non-acknowledged intermediaries exposes cross-border messages to various security and privacy concerns, triggering a demanding request for confidentiality, integrity, authentication, and non- repudiation. Such an uncompromising request for security is out of balance with the required low-latency property of cross-border transactions. A supplementary aspect is the need for compliance with country-defined regulations. The optimization objectives de- fined herein, comprising end-to-end latency minimization, end-to- end throughput maximization, end-to-end reliability maximiza- tion, and security level maximization, face multiple constraints, including compliance with international security regulations, flow scalability, data locality demand, and implementation complex- ity. More specifically, routing path selection, protocol encoding decision, cryptographic parameterization selection, and message transfer ordering mechanism are used as decision variables. Latency distribution, success rate of transmission, accuracy of anomaly detection, security risk related to attack surface extension, security risk associated with confidential data leakage, security risk inferred from privacy-preserving transformation, and compliance with regulation conformance act as evaluation metrics. A set of synthetic data traces of global routing, as well as real cross-border financial message logs, constitute the input dataset. Furthermore, the proposed optimization problem formulation is a surrogate and can be solved without low-data- count concerns.