This paper addresses the challenge of road network trajectory prediction in complex urban environments, where capturing both spatial and temporal dependencies is essential for accurate movement prediction. We propose a hybrid model that integrates Graph Neural Networks (GNNs) and Transformers to jointly model spatial relationships and temporal dynamics in road networks. The GNN component includes a fine-grained spatial and semantic encoder that captures road attributes such as connectivity, segment characteristics, and localized topological details, providing a deeper understanding of road network structure. The Transformer component models long-range temporal dependencies, complementing the spatial insights provided by the GNN. Furthermore, we introduce a Contextual Candidate Attention Module that leverages an attention mechanism to dynamically evaluate and prioritize candidate road segments, ensuring that the model’s predictions are contextually informed and semantically rich. Experiments on real-world datasets demonstrate that our model outperforms state-of-the-art methods.