Edge-Native AI: Optimizing Federated Learning for Privacy-Preserving IoT Data Mining

The rapid proliferation of Internet of Things (IoT) devices has led to unprecedented volumes of sensitive, high-velocity data being generated at the network edge. Conventional cloud-centric machine learning approaches struggle to meet the stringent privacy, latency, and bandwidth requirements of such environments, while also exposing raw data to centralized aggregation risks. Federated Learning (FL) has emerged as a promising privacy-preserving paradigm by enabling decentralized model training; however, existing FL frameworks are largely cloud-orchestrated and insufficiently optimized for heterogeneous, resource-constrained edge ecosystems. This article introduces the concept of Edge-Native AI, an architectural and algorithmic paradigm in which artificial intelligence models are designed from inception to operate natively within edge-centric IoT infrastructures. The study investigates how federated learning can be systematically optimized under an edge-native design to enhance privacy preservation, communication efficiency, and learning performance for IoT data mining tasks. We propose an edge-native federated learning framework that integrates adaptive client selection, communication-aware model updates, and lightweight privacy-enhancing mechanisms tailored to non-IID and dynamically evolving IoT data. Through analytical discussion and experimental evaluation grounded in recent advances in edge computing and privacy-preserving machine learning, the results demonstrate that edge-native optimization significantly reduces communication overhead while maintaining competitive model accuracy and stronger data locality guarantees. The findings highlight the practical and theoretical significance of Edge-Native AI as a foundational approach for scalable, privacy-preserving IoT data mining in next-generation intelligent systems.