Cross-Regional Adaptable ITS with Federated Learning: Enhancing Human-Centric Interaction and Technical Efficiency

Cross-regional traffic pattern heterogeneity and neglect of human needs limit ITS scalability. This paper proposes a cross-regional adaptable federated ITS that integrates human-centric design with technical optimization. The framework uses federated few-shot learning to adapt to new regions with minimal data, combined with dynamic sparsity to reduce communication overhead. A multimodal decision engine fuses traffic control logic, spoken language understanding for driver commands, and zero-shot recommendation for personalized services. Human factors are incorporated via VR-based interface design and cognitive state inference. Performance is evaluated using SBM-DEA to balance technical metrics (prediction accuracy, communication cost) and user-centric outcomes (satisfaction, cognitive load). Experiments show the framework outperforms baselines in cross-regional adaptation time (10.2 minutes vs. 45.3 minutes for FedAvg) and user satisfaction (89% vs. 65.7% for non-human-centric models). This work addresses ITS scalability challenges while prioritizing driver experience.