Time Series: Covid-19 should you be worried? What Machine Learning says.

Original article can be found here (source): Artificial Intelligence on Medium

(rights: source)

Time Series: Covid-19 should you be worried? What Machine Learning says.

According to data you should probably be …

The goal of this article is to both reassure but then also to raise the reader’s awareness of the gravity of the current situation. We will simply be analyzing the data and be doing predictions on the potential number of cases if no new measures are taken, through times series.

Some general figures

As of now (while writing),437 043 individuals were infected or are still infected by the coronavirus among which 19 645 deaths but also 111 861 fully recovered cases. Here is an overview of the situation (fig.1).

fig.1 from 25 March (rights: own image)

Here we show a higher view of the map for the coronavirus (fig.2):

fig.2 from late February (rights: source)

As you can see, the threat is pandemic and is already severe in many continents.

Quick Analysis:

  • The way people usually compute the mortality rate is computing the number of deaths over the number of cases, this gives a nearly 2%. However, when you compute the number of deaths over the number closed cases you would get a rough 4% rate (this constitutes an upper bound)
  • Unlike other viruses, Covid-19 is totally new, meaning that the population that we are working with is a relatively small sample. There is a huge variance on any sort of rate. Meaning that a sampled 4% rate could very well be actually 5% OR 3%.

Let’s study the evolution of cases in most viral countries so far. (fig.3)

fig.3 (rights: own image)

As you can see (fig.3), countries are in different phases of the virus. Indeed, China is in a stable situation in comparison to the UK that seems to have just stepped into the crisis. Notice that the US, for instance, is the country that has the most sudden increase over the past few days.

fig.4 (rights: own image)

Worldwide, seemingly the crisis just started. But again, now that the world has been warned, the curve perhaps would smooth up quicker.

Interval is taken from days where daily increase greater than 1k fig.5 (rights: own image)

Even though this is only one case, and this value might vary vastly among countries, it took approximately 25 days (fig.5) for China to go through the harshest increase. Considering that measures taken there are very much strict, we could consider the 25 days as a sort of lower-bound.

Forecasting (Take it with a grain of salt)

We will be using ARIMA model (fig.6) to forecast the number of cases in the near future.

fig.6 (rights: own image)

Here (fig.7) is what we got when training using China’s dataset, for the first 52 days.

fig.7 (rights: own image)

And then here is the short forecasting for China, US, UK, and France that we make:

(rights: own image)
(rights: own image)
(rights: own image)
(rights: own image)


In no mean am I saying that these increases will actually happen (and I really wish there was no increase in number of cases).

Note (for plots):

The forecasting should only be used when the period that we try to predict is within a specific phase. Meaning that you should not try to predict the evolution of the cases in 100 days and expect it to tone down when the country is still in a phase of increase. Plus we remind that the phase of increase (defined by at least 1000 new cases daily) is more or less 25 days long.


We see some very warning figures in European countries and the US that are forecasted to double in numbers after 5 days. Plus the rate of mortality that increases among youngsters due to the hospitals being very much overloaded. There are more than enough reasons to be worried, but again it is not time to panic, seeing that the situation is stabilizing in some countries (such as China, Japan, Korea), we do have hope for a tone down of the crisis. We do have hope for a normal upcoming summer, we will manage. Please hang in there! And contribute to the effort whenever you can! Through either practicing social distancing, or volunteering!