Privacy-Preserving Technologies in Edge Analytics: Architectures, Mechanisms, and Performance Frontiers

The shift of data analytics from centralized clouds to the network edge enhances latency and bandwidth efficiency, but it also raises significant privacy concerns. As sensitive biometric, behavioral, and operational data are gathered on resource-limited IoT devices, traditional security measures become outdated. This paper explores the main challenges in securing edge analytics, emphasizing the "Privacy-Utility Paradox" and the risks of inference attacks. It provides a detailed analysis of distributed learning architectures—Federated Learning (FL), Split Learning (SL), and Swarm Learning—and assesses cryptographic techniques such as Differential Privacy (DP), Homomorphic Encryption (HE), and Secure Multi-Party Computation (SMPC). Based on recent quantitative research, we propose mitigation strategies that balance accuracy, latency, and energy use, guiding the development of compliant and trustworthy edge intelligence systems.