Advanced computational models for urban traffic flow prediction: A comprehensive review and future directions

Traffic flow prediction is a fundamental task in intelligent transportation systems (ITS), supporting efficient mobility management and smart city development. In recent years, ITS research has rapidly progressed from traditional statistical models to advanced deep learning architectures, including convolutional, recurrent, graph-based, and attention-driven spatio-temporal networks. This article provides a comprehensive review of these approaches, categorizing them by methodological families, summarizing their strengths and limitations, and comparing their performance on widely used benchmarks. A particular emphasis is placed on federated learning, an emerging paradigm that enables collaborative model training across cities, operators, and edge devices without exposing sensitive data. We outline key application scenarios for federated traffic prediction, analyze technical challenges such as independent and identically distributed (IID) and non-IID data distributions, communication overheads, and privacy risks, and highlight representative solutions proposed in the recent literature. In addition, we compile a repository of publicly available datasets and summarize benchmark results to facilitate reproducibility and fair comparison. Finally, we identify open challenges and promising directions, including federated graph learning, explainable and trustworthy AI, and resource-aware deployment. This review aims to serve as a reference for researchers and practitioners, offering both a structured overview of the state-of-the-art and a roadmap for future advances in traffic flow prediction.