نورس العاشور

In fact, the accuracy of traffic flow prediction plays a vital role in the optimization of Intelligent Transportation Systems. In this regard, the traditional centralized approach deman‎ds the aggregation of the raw data, which in turn raises serious privacy issues. In this paper, a new traffic flow prediction framework with the ability to provide a high level of privacy has been proposed using the decentralized approach of Federated Learning with the assistance of a simulated multi-tiered network. To address the issue of the complexity of the spatio-temporal relationships between the traffic flows, a new streamlined deep learning model has been employed with the assistance of Graph Neural Networks an‎d Long Short-Term Memory units. Moreover, to provide the security requirements of the system, the model has been equipped with a robust security feature using the Differential Privacy approach with the assistance of a two-step gradient perturbation mechanism. The proposed model has been eva‎luated using the popular PeMS-BAY an‎d METR-LA traffic flow datasets. From the experimental results, it is evident that a privacy-utility tradeoff is achievable, with the ideal epsilon value being 1.0 for a suitable balance in the protection of identity inference an‎d precision in the modelʹs functionality. Moreover, the validation across datasets ensured the modelʹs ability to generalize across diverse urban environments, with performance remaining consistent with centralized benchmarks. Notably, the federated approach achieved a mean absolute error rate of 3.7221 mph on the PeMS-BAY dataset, thus demonstrating the ability for privacy-by-design without sacrificing utility in the process. This research, therefore, presents a secure framework for the development of future data-based mobility services, effectively addressing the need for precise traffic flow prediction while meeting strict regulatory an‎d ethical guidelines for data protection.

يادگيري فدرال , پيش‌بيني جريان ترافيك , حفظ حريم خصوصي , حريم خصوصي افتراقي , مدل‌سازي فضايي-زماني , سيستم‌هاي حمل و نقل هوشمند (ITS).

Federated learning , Traffic flow prediction , Privacy preservation , Differential Privacy , Spatiotemporal Modeling , Intelligent Transportation Systems (ITS).

Nawras AL Ashoor

Naser Mozayeni