<p>Graph Neural Networks (GNNs) have emerged as a powerful tool for traffic flow prediction, demonstrating significant advancements in modelling complex spatial-temporal dependencies in traffic networks. This survey presents a comprehensive review of GNN applications in traffic flow prediction from 2020 to 2024, offering unique insights through an extensive quantitative analysis. Unlike previous reviews, our work provides an end-to-end examination of the entire prediction pipeline, from data processing to model deployment, with a particular focus on recent advancements in graph construction methods, feature engineering and network architectures. The key contributions of this survey are threefold: (1) We present a comparative analysis of model performance across multiple datasets and prediction horizons, evaluating around 40 state-of-the-art models on five major public datasets, spanning short-term (15 and 30-min) and long-term (60-min) prediction horizons. (2) We systematically organize and summarize different graph construction methods, feature selection and fusion techniques, and various structural designs in GNN-based traffic prediction. This includes a comprehensive examination of static, adaptive, and dynamic graph constructions, multi-view and hypergraph approaches, as well as emerging trends such as physics-informed GNNs and hybrid architectures. (3) We offer a critical analysis of real-world implementation challenges, including scalability, computational efficiency, and strategies for handling data quality issues, alongside identifying promising future research directions. By providing this comprehensive, quantitative evaluation alongside a thorough review of recent advancements, our survey offers researchers and practitioners a clear understanding of the current state-of-the-art in GNN-based traffic prediction.</p>