Generalized Additive Model for traffic prediction

This repo contain 3 Jupyter Notebooks for trajectory prediction of pedestrains, based on velocity & direction. The results outperform many deep learning model, sometimes without any roadmap information.

Setup

The datasets files should be downloaded and extracted in the same directory as the notebooks. Namely, they are:

• Stanford Drone Dataset : Just the text files are needed, each inside its folder in a parent data folder, Download from here
• Intersection Drone Dataset: The csv files are needed inside indds parent folder. Download from here
• Argoverse 2 for Motion Forcasting: train and val files should be extracted directly next to the notebook Download from here

After that required python packages should be installed by:

pip install -r requirements.txt

Lastly, the notebook can be run directly cell-by-cell to check the results.

Cition info :

If you used this work in your porject, consider citing the following paper:

@article{yousif2024efficient,
  title={Efficient and Interpretable Traffic Destination Prediction using Explainable Boosting Machines},
  author={Yousif, Yasin and M{\"u}ller, J{\"o}rg},
  journal={arXiv preprint arXiv:2402.03457},
  year={2024}
}