DynST: Large-Scale Spatial-Temporal Dataset for Transferable Traffic Forecasting with Dynamic Road Networks

Zhiwei Zhang, Yanke Yu, Kelan Peng, Shaojun E, Fandong Meng, Jie Zhou, Wenjuan Han

Keywords: Traffic Forecasting; Transfer Learning; Spatial-Temporal Data Mining; Dataset;

TL;DR: We propose a large-scale dynamic road network dataset, named DynST, for transferable traffic forecasting.

Abstract: In real-world traffic networks, it is common to encounter a shortage of historical data in the target region. Researchers often address this issue through transfer learning. However, transfer learning tasks in traffic prediction currently lack dedicated datasets and instead rely on datasets designed for non-transfer prediction tasks. The major drawback of these existing datasets is the adoption of a fixed network topology to model the real world's road networks. This does not align with reality and limits the model's transferability. To tackle this issue, we propose DynST, a dataset specifically designed for transfer learning tasks in traffic prediction, with a massive data volume of 20.35 billion, spanning 20 years and 9 regions. The key feature of DynST is evolving dynamic road network topology, which reflects the evolution of real road networks. Moreover, to address the shortcomings of the distance-based adjacency generation algorithm, we introduce a novel tree-based algorithm. Extensive experiments demonstrate that the adoption of DynST as the source dataset can significantly enhance the performance of the target region. The comparative experiment also validates that our adjacency matrix generation algorithm can lead to improved prediction accuracy. We believe that DynST, with rich spatial variation information, will facilitate research in the field of transfer traffic prediction.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8583