Federated Learning for Sustainable Smart Mobility: Enhancing Privacy and Reducing Cloud Dependency
Andrei Toma, Radu‐Ioan Ciobanu, Ciprian Dobre, Gabriela Răducan · 2025
Federated learning (FL) is emerging as a sustainable approach in machine learning, enabling privacy-preserving, decentralized model training while reducing the energy and data transfer costs associated with centralized cloud-based computation. In the context of outdoor human mobility, where location data are sensitive, FL provides an efficient, low-carbon alternative for mobility prediction without requiring direct data sharing.This paper explores the application of FL for sustainable smart mobility, demonstrating its potential to optimize urban planning and intelligent transportation systems while preserving user privacy. By limiting data movement and leveraging edge computation, our FL-based approach significantly reduces cloud dependency, contributing to greener AI systems. Simulation results show that while FL exhibits a slight performance trade-off, it offers strong sustainability benefits by decreasing data center loads, network congestion, and computational overhead.We thus aim to explore the potential of federated learning for analyzing outdoor human mobility and to compare its outcomes with those of a traditional centralized machine learning method. We assess various model architectures, parameter update aggregation techniques, and the participation level of clients in the learning process. This research highlights the strengths and limitations of FL in maintaining data security while training models on sensitive user data in scenarios of outdoor human mobility.