Federated Learning for Privacy-Preserving Data Mining in Smart Cities

Abstract

The emergence of smart cities has revolutionized the collection and utilization of urban data to enhance public services, transportation systems, energy efficiency, and security. However, the proliferation of connected devices and sensors introduces critical challenges in data privacy and governance. Traditional centralized machine learning approaches require raw data aggregation, which risks breaching user privacy and data protection regulations. Federated Learning (FL) has emerged as a transformative solution, allowing collaborative model training across decentralized devices without sharing raw data. It elaborates on the architecture, workflow, and key algorithms used in FL and investigates use cases such as traffic flow prediction, energy demand forecasting, and public health monitoring. We also discuss the technical challenges, including data heterogeneity, communication overhead, and system robustness. Furthermore, the paper evaluates recent research outcomes and proposes future directions to enhance the scalability, trustworthiness, and real-time capability of FL systems in smart city infrastructures.

Keywords: Federated learning, smart cities, privacy, distributed data mining