Uncertainty-aware train delay prediction

This code accompanies our paper Quantifying the dynamic predictability of train delay with uncertainty-aware neural networks published in Transportation Research Part C: Emerging Technologies.

In this paper, we propose a framework for analyzing the dynamic predictability of train delays at varying horizons, and present an uncertainty-aware neural network approach that outperforms other methods.

Unfortunately, the dataset is not publicly available. With similar data, this code can be used to train uncertainty-aware neural networks and to evaluate the predictions. The data should contain observations of the current and final delay, in form of a csv file with the main columns train_id, obs_count, final_delay, current_delay, etc.

The following steps are executed:

1) Add features to the data (this takes a few minutes):

This will take the raw data and preprocess it, adding some columns with features to the data.

python add_features.py --inp_path path_to_data

2) Train model

The following command will train a Neural Network with aleatoric and epistemic uncertainty estimates, and save the model in a new folder within the trained_models directory. The flag -e determines the number of epochs that the model is trained.

python train.py -m nn -o out_dir_name -e 50

All code to train and test the neural network with aleatoric and epistemic uncertainty estimation is provided here.

3) Evaluate the results

We provide a script to run the model on test data and to compute the MSE, MAE, prediction interval width (as a metric for the precision of the uncertainty-enhanced predictions) and the likeliness of realization. This script evaluates all models in the specified folder.

python run.py -m model_folder_name

(here, model_folder_name would be the same as out_dir_name above)

4) Plot the results

Reproduce figures from our paper with the following command:

python plotting.py

References

Spanninger, T., Wiedemann, N., & Corman, F. (2024). Quantifying the dynamic predictability of train delay with uncertainty-aware neural networks. Transportation Research Part C: Emerging Technologies, 162, 104563.

@article{spanninger2024quantifying,
  title={Quantifying the dynamic predictability of train delay with uncertainty-aware neural networks},
  author={Spanninger, Thomas and Wiedemann, Nina and Corman, Francesco},
  journal={Transportation Research Part C: Emerging Technologies},
  volume={162},
  pages={104563},
  year={2024},
  publisher={Elsevier}
}