A COMPREHENSIVE FRAMEWORK FOR EDGE-ENABLED FEDERATED LEARNING IN IOT: STUDY OF DISTRIBUTED INTELLIGENCE, PRIVACY, SECURITY, AND COMMUNICATION EFFICIENCY

Abstract

The rapid growth of the Internet of Things (IoT) has generated large volumes of heterogeneous and decentralized data, much of which is privacy-sensitive. This creates significant challenges for the scalability, latency, and security of traditional cloud-based machine learning approaches. Edge computing and federated learning have emerged as effective solutions to these limitations by enabling distributed intelligence and collaborative model training directly at the network edge, while reducing the exposure of raw data. This study presents a comprehensive analysis of edge-enabled federated learning for IoT systems. It integrates architectural foundations, optimization algorithms, communication protocols, and privacy-preserving mechanisms into a unified framework. The study systematically examines major challenges in IoT federated learning, including statistical and system heterogeneity, communication bottlenecks, data quality limitations, and adversarial threats. It also discusses key solutions such as adaptive aggregation, secure aggregation, and privacy-preserving optimization. To extend the theoretical analysis, an empirical evaluation of two federated optimization algorithms, FedAvg and FedNova, was conducted under realistic edge-IoT communication constraints using a convolutional neural network and distributed client simulation. The experimental results show that aggregation normalization improves early-stage convergence stability. However, FedAvg and FedNova achieved comparable final accuracy and communication overhead under homogeneous conditions. The findings suggest that aggregation strategies alone are insufficient to significantly reduce communication cost. Therefore, integrated approaches combining adaptive optimization, model compression, and hierarchical coordination are required for more communication-efficient and scalable edge-enabled federated learning in IoT environments.