We model the number of deaths and severe cases by country for COVID-19 and make projections for the next 2 months. Our novel SELICRD model is an extended version of the common epidemiological SIR model for infectious disease progression. Briefly, the model considers country-specific values for population, hospital bed capacity, and R0 over time. Separate sub-models are used to estimate R0 and severe case counts.
This is a project done for a course at Stanford University (CS472: Data Science and AI for COVID-19) in partnership with Genentech, Inc. during the Spring 2020 quarter, in the midst of the 2020 COVID-19 pandemic. We are thankful for inspiration from models by Henri Froese (https://towardsdatascience.com/infectious-disease-modelling-fit-your-model-to-coronavirus-data-2568e672dbc7) and Michael Li (https://www.covidanalytics.io/DELPHI_documentation_pdf).
The final writeup can be accessed in the file COVID19_SELICRD_Predictions_Final_Writeup.pdf.
Data: Our data sources are enumerated in Section 3 of our report and within the individual scripts.
The data used for the most recent script run is contained within the /Data folder. This data was last updated on 6/10/2020.
Model Diagram: For those familiar with the SIR model, a model flow diagram for the SELICRD model is available in SELICRD_model_diagram.png.
Main Script: It is possible to make predictions for other regions and countries given data on population count, number of hospital beds, number of deaths, and number of severe (critical care/ICU) cases.
The script to run the SELICRD, time-changing R0, and SLSC models is full_model.ipynb and is written in Python 3. This script contains the full pipeline from modeling to validation and making projections.
Projections: Predictions for the number of severe cases for the next 2 months for chosen countries are available as CSV files in the folder /Projections.