Hybrid Transformer–Federated Learning Model for Secure Release Engineering in Global Payment Networks

Global payment networks encounter significant difficulties in administering safe software releases owing to diverse systems, sensitive client information, and escalating deployment difficulty. Conventional centralised learning methods pose privacy risks and do not effectively identify cross-silo patterns. This study introduces a Hybrid Transformer– Federated Learning model for secure release engineering, utilising Transformer encoders to extract multi-modal features from CI/CD logs, release notes, and transaction monitoring, while ensuring data privacy through federated optimisation, secure collection, and differential privacy. Solid methods, such as Krum and reduced mean aggregation, counteract hostile client attacks and abnormal updates. Experimental assessment utilising datasets from Kaggle reveals exceptional performance with an Accuracy of 0.95, Accuracy of 0.93, Recall of 0.95, F1-score of 0.94, and AUC of 0.96, surpassing baseline models such as Random Forest, LSTM, and FedAvg Transformer. The proposed architecture facilitates cross-silo knowledge sharing, secure risk prediction for releases, and anomaly detection, all while preserving client data security. The model surpasses existing methods in predicting accuracy, robustness, and privacy protection, providing practical utility in the deployment pipelines of global payment networks. This study establishes a basis for privacy-conscious, scalable, and dependable release engineering systems within distributed financial contexts.