COVID-19 is a hot topic these days. Healthcare workers are the first line of defense. If you are in IT you are part of the fight against the virus. I thought I should do my part and implement a method to forecast coronavirus growth and dates when the number of infections could stabilize.
What it does
The forecast is calculated daily based on the last available data, using logistic and Hill functions. Backtesting is done by forecasting for 5 days in the past. We provide a forecast for the next 20 days, along with backtesting data. Forecast helps to evaluate the situation and check how the country managing the virus outbreak.
How we built it
Challenges we ran into
Initially, we implemented a forecast using a logistic function, but this approach wasn't accurate enough. In multiple scenarios, we saw that there could be a sudden increase or drop in new virus cases. It proved that the second approach we implemented with the Hill equation (another kind of logistic growth) could handle forecast with high ups/downs better.
Accomplishments that we're proud of
Application is live and we are receiving around 1000 API calls daily. Getting positive feedback through social networks.
What we learned
We learned that virus growth tends to follow the logistic curve. This made it possible to build a forecast app. Obviously the forecast could be adjusted if there will be sudden updates in new virus cases. Still, it gives a better understanding of the situation with the virus in each selected country.
What's next for COVID-19 Forecasting
We are planning to add a forecast correlation with the number of daily patient tests and the number of patients recovered daily.
Source code: https://github.com/katanaml/covid19
Live app: https://app.katanaml.io/covid19/
Technical article I: https://bit.ly/2XHjL4P
Technical article II: https://bit.ly/2RMB14V