Skip to content

With this repository, I derive the time-dependent R0 coefficient of the COVID-19 with the Unscented Kalman Filter from the data gathered by John Hopkins assuming the SEIR model.

License

Notifications You must be signed in to change notification settings

jpcorb20/covid19-transmission-ukf

Repository files navigation

Derivation of R0(t) of many countries with SEIR model and the Unscented Kalman Filter.

Inspired by Viktor Kärnstrand's work on Medium: https://towardsdatascience.com/estimating-the-effect-of-social-distancing-in-sweden-c6c1e606c8f9.

Derived R0 with french country names:

PLOT OF R0

We observed the effect of increasing the measurement rate with the peak for all country between 0 and 20 days. Afterwards, the drop in transmission rate below the 5.7 days in recent literature is due to the confinement.

I, E and R populations derived for Canada and Italy with error margins of 3 standard deviations.

PLOT OF CANADA

PLOT OF CANADA

All examples have predictions for 5 days, which explained the rise in uncertainty in the last few days. All curves have been smoothed with a window of 10 days for plots and UKF (to stabilize the algorithm). Also, all UKF parameters have been fine-tune by hand.

These plots were done for all dates until May 11th 2020.

Once, these days passed, the plots are :

PLOT OF CANADA

PLOT OF CANADA

We see that our predictions were quite good actually for the Infected. For the Recovered, we fixed the recovery rate to the literature value, which might not be the best. We might modelised it also like the R0.

Last update May 16th 2020.

About

With this repository, I derive the time-dependent R0 coefficient of the COVID-19 with the Unscented Kalman Filter from the data gathered by John Hopkins assuming the SEIR model.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages