Towards an Original Privacy-Preserving Federated Anomaly Detection Architecture for Distributed Nigerian E-Government Systems

Nigerian e-government systems are distributed across multiple agencies with data residency and sovereignty constraints that prevent centralised security monitoring, creating anomaly detection blind spots that attackers exploit to persist across agency boundaries. This study develops an original privacy-preserving federated anomaly detection architecture for distributed Nigerian e-government systems. The architecture, designated FedGovAnomaly, integrates five original technical contributions: a federated isolation forest variant that trains local anomaly detectors at each participating agency and aggregates them using a weighted model fusion protocol without exchanging raw event logs; a secure aggregation protocol based on homomorphic encryption that protects local model parameters during aggregation; a cross-agency alert correlation module that identifies distributed multi-step attack patterns using a graph-based incident linkage approach; an adaptive threshold calibration mechanism accounting for agency-specific traffic baselines; and a formal privacy guarantee based on differential privacy analysis. FedGovAnomaly was evaluated on synthetic distributed attack scenarios derived from Nigerian government network incident reports and compared against centralised and non-federated baselines. The federated architecture achieved 93.1 percent anomaly detection rate with 5.8 percent false positive rate, compared to a centralised baseline of 96.2 percent detection and 4.1 percent false positive. Privacy analysis confirms formal epsilon-differential privacy guarantees at epsilon of 4.0. The framework represents an original contribution to federated security analytics applicable to distributed e-government architectures globally.

Keywords: federated learning, anomaly detection, e-government security, privacy-preserving AI, Nigeria