Federated Deep Learning Models for Privacy Preserving Analytics in IoT Enabled Smart Cities

The rapid growth of IoT-enabled smart cities has led to the generation of massive volumes of real-time data from interconnected sensors monitoring traffic, waste management, energy usage, and environmental conditions. While this data offers significant potential for intelligent urban analytics, it also introduces critical concerns around data privacy, centralized storage vulnerabilities, and compliance with data protection regulations. To address these challenges, this paper presents a novel Federated Deep Learning (FDL) framework designed for privacy-preserving analytics in smart city environments. The proposed architecture, termed Hybrid Secure-FedNet, enables decentralized training across distributed IoT nodes without transferring raw data. It integrates lightweight convolutional and recurrent neural networks to handle spatial-temporal sensor data while incorporating Differential Privacy (DP) and Homomorphic Encryption (HE) techniques to safeguard model updates during communication. Experiments conducted on multiple open smart city datasets, including air quality and traffic data, demonstrate that our approach achieves comparable or higher accuracy (up to 93.2%) than centralized models, The proposed model is scalable, resilient, and well-suited for real-world deployment in data-sensitive smart urban infrastructures