Federated-Learning based Deep Learning framework for Data privacy in Smart city applications

The rapid increase in data within smart city infrastructures—ranging from traffic management to health monitoring systems—presents significant opportunities for machine learning innovations. However, centralising such varied and sensitive information poses substantial challenges concerning data privacy, regulatory adherence, and scalability of systems. This paper introduces a secure and scalable Federated learning (FL) framework designed specifically for smart city settings, facilitating decentralised model training while maintaining localised data integrity and confidentiality. The framework incorporates essential technologies such as differential privacy measures, secure aggregation methods, and edge device optimisation to ensure reliable model performance under practical conditions. Implemented using TensorFlow with simulated smart city datasets, our evaluation covers critical metrics like training accuracy, communication expenditure, latency periods, and model convergence rates. Experimental findings indicate that the proposed FL framework delivers high predictive accuracy (94.3%), alongside markedly reduced bandwidth usage while upholding robust privacy protections. This research offers a viable architecture for forthcoming smart cities that strikes an optimal balance between efficient data utilisation and safeguarding citizen rights.