Abstract—Precise traffic prediction is paramount for optimised city mobility and smart transportation systems. Current models, though, are based on deep multi-layer structures, which are resource exhaustive and limit real-time implementation. This paper introduces Graph-Convolutional Temporal Transformer (GCTformer), a light spatio-temporal model that combines a one-layer Graph Convolutional Network (GCN) with a Transformer encoder. The innovation of GCTformer is that it can learn the adjacent matrix dynamically so that it accommodates adaptive spatial dependency modeling regardless of static or pre-designed graphs. Comprehensive experiments on three benchmark datasets Los Angeles Metropolitan Region (METR-LA), San Francisco Bay Area (PEMS-BAY), and a traffic speed dataset for the Northeast–Beijing (NE-BJ) Traffic Dataset demonstrate that GCTformer achieves accuracy equal to or superior to state-of-the-art methods while saving parameters by 30–40%. When compared to multi-layer GCN-based models, GCTformer shows resistance to long-term forecasting, reducing Mean Absolute Percentage Error (MAPE) by up to 2.3% at Horizon-12. The results establish GCTformer as a practical and efficient solution for real-time traffic forecasting in intelligent transportation systems. The implementation is available at: https://github.com/imagrawalpalak/GCTformer
 
Keywords—traffic prediction, spatial-temporal modeling, Graph Convolutional Networks (GCN), transformer models, hybrid architecture, short-term prediction, intelligent transportation systems


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).