Efficient Traffic Prediction at Scale: A Systematic Study of STGCN Architectural Depth
Published in Accepted to the 2026 IEEE 29th International Conference on Intelligent Transportation Systems (ITSC), Naples, Italy, 2026
We ask whether spatio-temporal graph neural networks for traffic prediction are over-parameterized, using STGCN as a test case. Across four traffic datasets, a single-block STGCN matches or beats the standard 2-block design at a fraction of the computational cost.
Recommended citation: Lone, Soban Nasir, Mohamed Abouelela, Taeyoung Yu, Jiwon Kim, and Constantinos Antoniou. "Efficient Traffic Prediction at Scale: A Systematic Study of STGCN Architectural Depth." Accepted to the 2026 IEEE 29th International Conference on Intelligent Transportation Systems (ITSC), Naples, Italy. arXiv preprint arXiv:2606.09539 (2026).
Download Paper
