A federated learning model enables secure data sharing across regions in digital governance
TL;DR
A governance-constrained federated learning model for cross-regional data sharing that enables the global model to adjust aggregation weights, privacy intensity, and transmission frequency according to node heterogeneity, data sensitivity, and governance authorization status is developed.
Abstract
To support secure collaboration among heterogeneous regional governance nodes, this study develops a governance-constrained federated learning model for cross-regional data sharing. Different from conventional federated learning security frameworks that separately deploy privacy protection, encryption, compression, and access control, the proposed model couples node contribution estimation, privacy-risk-aware perturbation, communication-load scheduling, and compliance decision constraints within one adaptive training process. This design enables the global model to adjust aggregation weights, privacy intensity, and transmission frequency according to node heterogeneity, data sensitivity, and governance authorization status. Experimental evaluation was conducted using multiple datasets and distributed training environments, and the proposed model was compared with FedAvg, FedProx, SCAFFOLD, FedNova, DP-FedAvg, and QSGD-FedAvg. The improved model achieved an accuracy of 92.6%, an F1 score of 91.8%, and an RMSE of 0.109. The privacy leakage risk decreased to 0.087, the communication volume was reduced to 1.5 MB, the total training time was 131 s, the stability metric reached 0.95, and the compliance rate increased to 96.8%.