Urban mobility has become a pressing challenge for societies worldwide, as traffic congestion, environmental degradation, and inequitable access to transportation increasingly affect economic growth and quality of life. Predicting traffic flow accurately is a cornerstone of intelligent transportation systems, yet the problem remains complex due to the highly dynamic, nonlinear, and contextdependent nature of urban mobility. Traditional machine learning and deep learning approaches have made substantial progress by capturing spatialtemporal correlations within traffic networks, but they often fail to generalize across cities, handle data sparsity, or incorporate heterogeneous information sources. This paper proposes a metalearning framework that integrates multisource spatiotemporal data—including traffic counts, weather conditions, points of interest, and urban events—into a unified predictive model. By combining metalearning with multisource fusion strategies, the framework is capable of learning transferable knowledge across urban contexts while adapting quickly to new environments with limited data. The study not only advances the technical methodology for traffic flow forecasting but also situates the discussion within broader social and policy perspectives, emphasizing fairness, sustainability, and realworld applicability. Through extensive evaluation on multicity datasets, the proposed framework demonstrates superior adaptability and interpretability compared to established baselines, paving the way for more equitable and intelligent transportation systems.
Geng et al. (Mon,) studied this question.