Urban traffic forecasting often requires data from multiple stakeholders, yet raw data cannot be freely shared because of privacy or commercial constraints, motivating federated spatio‑temporal approaches. In a federated setting each client observes traffic on its own sensor subgraph with a distinct spatial topology and temporal dynamics, which creates strong heterogeneity across clients. Existing federated spatio‑temporal methods mainly rely on model parameter aggregation and lack mechanisms to recover spatial dependencies across client boundaries, leading to two major issues: parameter aggregation over heterogeneous graph domains dilutes client‑specific representations, and road‑network partitioning breaks the propagation of traffic dynamics at client borders. To tackle these challenges we introduce FedeRICo, a federated traffic forecasting framework that combines gradient‑level collaboration with boundary‑aware residual communication. FedeRICo uses a dual‑branch forecasting architecture: a globally guided branch captures transferable forecasting structures, while a private residual branch preserves client‑specific corrections and incorporates boundary residual signals. The global branch is coordinated via gradient alignment across all clients, enabling collaborative optimization without destructive parameter interference. To restore cross‑client spatial dependencies, boundary messages are extracted through a trend‑residual decomposition that suppresses periodic patterns and transmits only transient spatio‑temporal residual signals between physically adjacent clients. Experiments on four real‑world traffic forecasting benchmarks demonstrate that FedeRICo consistently outperforms state‑of‑the‑art federated spatio‑temporal baselines while keeping training runtime competitive.
Review