ZERO-TRUST ACCESS CONTROL ARCHITECTURE FOR PRIVACYPRESERVING MACHINE LEARNING IN SMART EDGE NETWORKS

Abstract

The rapid adoption of Internet of Things (IoT) and edge computing has enabled real-time intelligence in smart environments, yet it has also introduced severe privacy, trust, and security challenges. Traditional perimeter-based security models are insufficient for distributed learning systems, where data and computation occur across heterogeneous and potentially untrusted nodes. This research introduces a ZeroTrust Access Control Architecture for Privacy-Preserving Machine Learning in Smart Edge Networks, designed to ensure secure collaboration among decentralized edge devices without exposing sensitive data. The proposed architecture integrates Zero-Trust security principles, federated machine learning, and fine-grained policy-driven authentication mechanisms to verify every device, user, and service before each interaction. A multi-layer encryption and identity-verification framework enforces continuous trust evaluation, while secure aggregation and differential privacy techniques protect model updates from inference attacks during distributed training. The architecture also incorporates context-aware access monitoring and anomaly detection to mitigate insider attacks, spoofing, and unauthorized model manipulation. Experimental results demonstrate that the proposed system significantly enhances data confidentiality, model integrity, and resilience against cyber threats, while maintaining low latency and computational overhead suitable for resource-constrained edge environments. This Zero-Trust-driven privacy-preserving framework provides a scalable and secure foundation for deploying machine learning in smart networks such as smart cities, healthcare, industrial IoT, and intelligent transportation systems.