Time-Series Forecasting for Humanitarian Aid - model

In this repository is possible to find the code built for the purpose of this project. The experiments conducted and the output were driven by the data gathered, through the use of two main technologies:

• ARIMA - Autoregressive Integrated Moving Average: a statistical model which perfectly fits for time series analysis and predictions of futre points (forecasting)
• TFT - Temporal Fusion Transformer: attention-based Deep Neural Network, optimized for multivariate time series forecasting.

To use our model, install the requirements and clone the dataset repositories:

git clone https://github.com/PoliTO-ADSP-United-Nations-Project/humanitarian_aid_dataset

and model repositories:

git clone https://github.com/PoliTO-ADSP-United-Nations-Project/tsf_model

Execute with Jupyter notebook:

You can find a complete Jupyter notebook file that shows how run both the models.
There is the possibility to open it through Google Colab.

Execute in local:

As an alternative, you can use the operating system terminal to execute the code in this way:

Create the dataset from scratch:

pip install -r "./humanitarian_aid_dataset/requirements.txt"
python ./humanitarian_aid_dataset/main.py

or download it from Figshare.

Get mandatory packages:

pip install -r "./tsf_model/requirements.txt"

Run ARIMA Model:

python ./tsf_model/statistical_models.py --dataset_dir="/content/final_dataset.csv" --destination_country="ITA" --model_name="ARIMA"

Run TFT Model:

python ./tsf_model/tft_model.py --dataset_dir="/content/Final_TFT.csv"

---

Please, refere to the official paper for further information.

References

If you find this work useful, pleace cite the paper above as:

@article{TSF_4_ha,
  title={Time series forecasting for humanitarian aid},
  author={Bergadano Lorenzo, Frigiola Arcangelo, Mantegna Giovanni, Scriotino Giovanni, Zingarelli Valerio},
  year={2023}
}