Federated Learning for Privacy-Preserving Smart Cities: A Secure and Scalable Machine Learning Framework

The exponential growth of data in smart city infrastructures—from traffic systems to health monitoring and surveillance—has created unprecedented opportunities for machine learning applications. However, centralizing such diverse and sensitive data introduces serious challenges related to data privacy, regulatory compliance, and system scalability. In this paper, we propose a secure and scalable federated learning (FL) framework tailored for smart city environments, enabling decentralized model training while preserving data locality and privacy. The framework integrates key technologies including differential privacy, secure aggregation, and edge device optimization to ensure robust model performance and security under real-world conditions. The framework is implemented and simulated using TensorFlow with synthetic smart city data streams, evaluating the system across key metrics such as training accuracy, communication cost, latency, and model convergence. Our experimental results show that the proposed FL framework achieves high prediction accuracy (94.3%) with significantly reduced bandwidth consumption and strong privacy guarantees. This work contributes a deployable architecture for future smart cities, offering an effective balance between intelligent data use and citizen data rights.