SUNT - Salvador Urban Network Transportation

🗺️ Overview

First and foremost, thanks for your interest in using our dataset, referred to as Salvador Urban Network Transportation (SUNT), and benchmarks.

Here you will find all codes, models, and data used in the manuscript “Salvador Urban Network Transportation (SUNT): A Spatiotemporal dataset on the public transportation”.

More details about our data, codes, and models are shown in each folder.

The following figure illustrates our dataset is geographically distributed in Salvador (Brazil).

On the behalf of all of the authors, we appreciate your interest in our data, code, and models, and hope they are useful to your research.

---

🗂️ Repository structure

The organization of this repository is:

• data : raw and graph-based data
• data_design : contains source codes developed to create our datasets and train learning models
• docs: database documentation
• images: plots with statistics of the dataset attributes
• integration: example of integrating other databases with the dataset
• models : contains frozen models used to predict different tasks using SUNT
• outputs/: weigths and results
• stats: notebook with data statistics.

---

💾 Dataset

The full dataset is stored on the Mendelay Data.

• RAW
  • AVL (Automatic Vehicle Location)
  • Automatic Fare Collection (AFC)
  • General Transit Feed Specification (GTFS)
  • Local Trip Information (LTI)
• Built
  • Boarding Hugging Face
  • Alighting Hugging Face
  • Origin-Destination (OD) Hugging Face

More dataset info, see dataset doc.

🚀 Practical Examples and Tutorial

To help users get started with SUNT, we provide practical and easy-to-run examples for data visualization and passenger-demand forecasting:

• SUNT Examples: Python scripts covering time-series and graph visualization, as well as forecasting with SARIMA, LSTM, GRU, Transformers, Chronos, GCN, T-GCN, and T-GAT.
• WCCI/IJCNN 2026 Tutorial — From Stops to Graphs: tutorial materials combining theory and hands-on examples for modeling public transportation data with time-series models and Graph Neural Networks.

The examples include instructions for installing the required dependencies, downloading the data, running the experiments, and reproducing the generated outputs.

📝 Citation

@article{Ferreira2025,
  author    = {Ferreira, Marcos V. and Souza, Matheus and Rios, Tatiane N. and 
               Fernandes, Islame F. C. and Nery, Jorge and Gama, Jo{\~a}o and 
               Bifet, Albert and Rios, Ricardo A.},
  title     = {Salvador Urban Network Transportation (SUNT): A Landmark Spatiotemporal Dataset for Public Transportation},
  journal   = {Scientific Data},
  year      = {2025},
  volume    = {12},
  number    = {1},
  pages     = {1320},
  doi       = {10.1038/s41597-025-05674-6},
  url       = {https://doi.org/10.1038/s41597-025-05674-6},
  issn      = {2052-4463},
}

📃 License

This project is licensed under the CC BY 4.0 License.