Privacy-Preserving Federated Learning Architecture for Cross-Domain IoT Threat Detection and Compliance

The rapid expansion of Internet of Things (IoT) ecosystems across healthcare, industry, smart cities, and critical infrastructure has amplified both the value of distributed telemetry and the security challenges associated with its use. Traditional centralised machine learning approaches, although effective, are increasingly incompatible with privacy regulations, data sovereignty constraints, and the risks inherent in aggregating sensitive cross-domain IoT data. This paper presents the Privacy-Preserving Federated Learning Architecture for Cross-Domain IoT Threat Detection and Compliance (PP-FL-IoT), a unified framework that integrates federated learning with differential privacy, homomorphic encryption, secure multiparty computation, Zero Trust security, and compliance-aware governance. Using real-world datasets and synthetic multi-domain simulations, the architecture demonstrates high threat detection accuracy that approaches centralised models while significantly reducing privacy leakage and improving adversarial robustness. Evaluation results show resilience against gradient inversion, membership inference, and poisoning attacks, alongside low communication overhead and practical detection latency suitable for time-sensitive IoT environments. Furthermore, PP-FL-IoT aligns with regulatory standards such as GDPR, HIPAA, ISO 27001, and NIST SP 800-53 through embedded governance controls and auditability. The study highlights how multi-layered privacy-preserving federated learning can enable secure, collaborative IoT threat intelligence without compromising data sovereignty or compliance obligations.