Securing IoT in Smart Cities with Federated Learning and Adaptive Clustering via FedAC Algorithm

Abstract

With the increasing number of IoT devices in smart city infrastructures, there is a growing need for advanced security measures to protect against new cyber threats. This research article presents the Federated Adaptive Clustering (FedAC) algorithm, a novel method designed to improve IoT-based smart city security through federated learning and privacy preservation. FedAC uses an adaptive clustering technique to group edge devices based on data similarity and computational power, optimizing local training and minimizing communication overhead. The technique guarantees robust data privacy by integrating differential privacy and safe multi-party computation within each cluster. The dynamic re-clustering feature adapts to changing data distributions and device availability, maintaining high model performance and efficiency. Experimental results show that FedAC achieves a 92.5% accuracy on real-world IoT datasets, while reducing communication overhead by 35% compared to traditional federated learning algorithms. Privacy loss is kept minimal with a privacy budget (epsilon) of 1.0, and computational efficiency is improved by 25% in convergence time. These findings highlight the potential of FedAC in strengthening smart city security, offering a scalable and resilient federated learning framework suited for the complexities of IoT environments.