Our proposed methodology has two main modules, see the graphical workflow in the below Figure:
(a)
(b)
Topological LSTM (a) RNN architecture. (b) Our proposed methodology
This package includes the source codes and datasets used in this research project. We encourage the reader to review the submitted paper: Covid-19 Forecasting via Deep Learning and Topological Data Analysis, and its supplementary material.
Our experiments have been carried out using collected data of California and Washington states. Particularly, our methodology produces daily COVID-19 progression and hospitalization forecasts at county-level resolution.
The complete software list and requirements are included in the file "Requirements.md".
The datasets for this research project are obtained from the below websites and repository:
- https://www.ncei.noaa.gov/
- https://github.com/CSSEGISandData/COVID-19
- https://midasnetwork.us/covid-19/
Please find the significance of each file and directory below:
- Datasets Directory: it contains all the publicly available datasets required for this research project.
- ChangeFormat_Hospitalizations_XXXX.R: These Scripts change the original format of county for each state data to our format for the deep learning model. The extracted datasets are stored in the CSV directory.
- LSTM_TDA_XXXX.py: This script fits a LSTM model on Covid data using files available in the InputLSTM directory and saves the forecastings into the Saved directory
- ExtractWeekly_Features_Washington: This Script changes the original format of county for each state data to our own format for the deep learning model.
- Dynamic Network.py: This script builds the Dynamic Network and extract topological summaries, and saves these results in a CSV file.