A comprehensive review of traffic flow forecasting based on deep learning

Traffic flow forecasting plays a pivotal role in Intelligent Transportation Systems (ITS), underpinning efforts to alleviate congestion, enhance mobility, and support sustainable urban development. The increasing complexity of urban traffic, driven by rapid urbanization and vehicle proliferation, necessitates accurate and adaptive predictive models. Recent advances in deep learning have significantly reshaped the landscape of traffic forecasting, enabling more effective modeling of spatiotemporal dependencies, nonlinear behaviors, and multi-scale patterns. This survey not only presents a comprehensive review of deep learning-based traffic flow forecasting but also differentiates itself from prior reviews by systematically comparing node-level and network-level paradigms under a unified framework, and by incorporating the emerging role of large language models (LLMs) in traffic prediction. Moreover, we propose novel research directions such as multimodal data fusion, dynamic graph learning, and explainable forecasting frameworks, which extend beyond traditional surveys. Overall, this work provides not just a structured overview but also a forward-looking roadmap that highlights underexplored yet promising directions for next-generation traffic forecasting systems. © 2025 Elsevier B.V.