On 26th April, Data Driven Innovation lab (DDI) at Singapore University of Technology and Design (SUTD) came out with an implementation of the popular SIR model and tried to predict the Covid19 peak dates for various countries including India. They predicted that India had already peaked on 20th April and 97% of the expected cases would have been recovered by May 25th. We were curious to verify these dates and built a methodology to validate these predictions.

Our method has a limitation that it can confirm the peak only 3 days after the actual peak date. After running our calculations, we found that the predictions are incorrect for most of the countries that are still in the rising half of the epidemic, where new highs are still being reported. The details of our analysis are presented here and our calculations and code is available in this notebook. The Author has used the popular SIR model but the inputs to the model have to be given with due understanding of the underlying dynamics of the population to get a good prediction.

It is still not confirmed that India has peaked. The actual date does not appear to be any time soon as per our calculations. So it is highly unlikely that 97% of expected cases would end by the forecast date for countries such as India, Pakistan, Saudi Arabia, UAE.. Further we predict that the daily new cases will never become zero as predicted by this paper but will leave behind an asymptote.

Results

First Updated: May 5, 2020. We plan to update these results every few days. So please check back for any changes.

Last Updated: May 19, 2020. Russia peaked on May 11th. However it’s death chart has not yet peaked. We have added a second chart for the country. Other charts have not been updated as they have not moved categories.

India

Here is a comparison of peak dates as per Singapore University’s paper and our study.

The following table shows a list of countries which haven’t hit the peak yet:

Country SUTD Our verification
Afghanistan Apr 29 not peaked
Bangladesh Apr 23 not peaked
Brazil Apr 21 not peaked
Chile Apr 16 not peaked
India Apr 20 not peaked
Indonesia Apr 20 not peaked
Mexico May 01 not peaked
Pakistan Apr 27 not peaked
Peru Apr 18 not peaked
Russia Apr 24 not peaked
Saudi Arabia Apr 27 not peaked
United Arab Emirates Apr 17 not peaked

This table shows countries that have already hit the peak number of new infections in a day:

Country SUTD Our verification
Australia Mar 27 Mar 28
Austria Mar 26 Mar 28
China Feb 08 Feb 13
Denmark Apr 06 Apr 07
Finland Apr 11 Apr 09
France Apr 03 Apr 02
Germany Apr 01 Apr 03
Greece Mar 30 Apr 02
Hungary Apr 15 Apr 12
Iran Apr 01 Apr 02
Iraq Apr 04 Apr 08
Israel Apr 04 Apr 04
Italy Mar 29 Mar 26
Malaysia Mar 31 Apr 04
Netherlands Apr 08 Apr 12
Norway Mar 27 Mar 29
Philippines Apr 07 Apr 03
Portugal Apr 06 Apr 04
South Korea Mar 02 Mar 03
Spain Apr 02 Apr 01
Sweden Apr 20 Apr 25
Switzerland Mar 29 Mar 30
Thailand Mar 28 Apr 01
United Kingdom Apr 12 Apr 15
United States Apr 10 Apr 10

Logic

We use basic calculus to identify the peak. A peak on the daily epidemic curve (i.e. new patients per day) is a maximum. To identify maxima and minima on a curve, we use the fact that first order derivative becomes zero at that point. So we calculate change in daily cases which denotes the slope of the curve and identify dates where this curve becomes 0. Each point is a potential peak.

However, we find multiple such points since the derivative becomes 0 whenever the curve is parallel to x-axis. To further verify that it is a true peak, we cross examine the second order derivative. At maximums, the second derivative, i.e. change in slope, should be negative because the slope changes from positive to negative at a peak.

Derivative Diagram

Graphs

Following is the list of charts for the countries mentioned above. For each country, there are two timeseries charts that have been plotted. First is the percentile epidemic curve, i.e. percentile of frequency of new patients per day. Second denotes the first order derivative of the epidemic curve, i.e. the change in daily cases. It is useful for finding peaks, which are marked by the vertical blue lines in the charts.

 

India
Back

United Arab Emirates
Back

Russia
Back

Russia1905
Back

Peru
Back

Mexico
Back

Pakistan
Back

Bangladesh
Back

Chile
Back

Saudi Arabia
Back

Brazil
Back

Afghanistan
Back

Indonesia
Back

China
Back

South Korea
Back

United States
Back

United Kingdom
Back

Italy
Back

Spain
Back

Germany
Back

France
Back

Portugal
Back

Australia
Back

Austria
Back

Denmark
Back

Finland
Back

Greece
Back

Hungary
Back

Iran
Back

Iraq
Back

Israel
Back

Malaysia
Back

Netherlands
Back

Norway
Back

Philippines
Back

Sweden
Back

Switzerland
Back

Thailand
Back