Machine Learning aided Epidemiology: COVID-19 Global quarantine strength and Covid spread parameter evolution

The quarantine strength function and the effective reproduction variation in several countries is estimated. The method followed is based on augmentation of the standard SIR epidemiological model with machine learning. Our model is universally applicable making it a flexible and powerful tool to analyze and compare the efficacy of government measures in curtailing virus spread in different countries.

Neural network schematic.
Illustration of the neural network architecture used to encode the quarantine strength function, Q(t). The video shows the learning of the neural network based on the infected and recovered case count data for UK, shown for limited data for demonstration purposes. The neural network learns the transition as the infected case count data changes regime from exponential to linear.
UK - Q function.
UK - Covid spread parameter.
Quarantine strength function. Q(t) and the Covid spread parameter, Cp(t) evolution for UK. The transition from the red to blue regionis when Cp becomes less than 1, indicating halting of the infection spread. Both of these quantities were obtained purely from the COVID-19 data, without any reliance on previous epidemics. Q(t) gives an indication of the increasing quarantine, lockdown and testing measures imposed in UK with time. Cp(t) evolution in UK shows that it took more than a month (32 - 33 days) to bring the Covid spread parameter down from >1 to <1 and thus halt infection spread.

Europe

France Germany Belgium Norway Denmark Turkey Sweden Serbia Czech Republic Russia Spain Italy UK Moldova North Macedonia Belarus Ukraine Netherlands Portugal Poland Romania Switzerland Ireland

Asia

India South Korea China Philippines Iran Pakistan Saudi Arabia Bangladesh Qatar Iraq Kazakhstan Kuwait Israel Japan Singapore Malaysia Afghanistan Bahrain Oman Armenia

North America

USA Mexico Canada Panama El Salvador

South America

Colombia Argentina Brazil Chile Peru Bolivia Ecuador Honduras Guatemala Dominican Republic

Africa

Egypt Algeria Morocco Cameroon Côte d'Ivoire Sudan Nigeria South Africa

Oceania

Australia

Clickable World Map

The regions for which our code is applied currently is shown in the above boxes. Alternatively, these regions can also be clicked on the map below to visualize the results.

Code and Data

The basic code is uploaded here. The code uses the method of universal ODE's developed here.

Data is collected from Center for Systems Science and Engineering (CSSE) at Johns Hopkins University.

Contributors