Next Generation Intelligent Transportation through Disaster Aware Privacy Preserving Federated Learning

The growing challenges posed by rapid urbanization and increased vehicle density have highlighted the need for precise and privacy-preserving traffic flow prediction models. Traditional centralized learning methods involve aggregating privacy-sensitive traffic flow data, which poses severe privacy, scalability, and governance issues. To overcome these issues, this paper presents a disaster-aware privacy-preserving federated learning (PP-FL) solution for multi-city traffic flow prediction in heterogeneous and non-IID data settings. Each city separately trains a local deep learning model using local traffic, environmental, and disaster-related data, while only privacy-protected model updates are exchanged with a central aggregator. The proposed solution leverages secure aggregation and differential privacy to ensure data privacy and facilitate efficient cross-city knowledge sharing. Extensive experiments on multi-modal urban traffic datasets, including junction-level surveillance traffic data, highway traffic time-series data, and disaster-influenced traffic data, validate that the proposed solution outperforms state-of-the-art centralized and federated learning baselines by up to ۸-۱۲% in RMSE error while substantially mitigating privacy leakage. The results also validate enhanced anomaly detection and disaster-caused traffic disruption prediction performance across cities with limited local traffic data. The obtained results validate the effectiveness of the proposed PP-FL framework as a scalable, secure, and efficient paradigm for next-generation intelligent transportation systems.

کلیدواژه ها:

Secure aggregation ، Scalability ، Secure data ، Efficient System ، Dataset