SAC Predictive Forecasting – The Corona Virus Test

A recent blog, on how easy it can be to run predictive scenarios and apply them to our forecast planning really, got me going on this subject. One thing led to another and in no time, here I am, trying it out on the Corona Virus data we have for England. It has been very interesting and thought provoking. I hope you find this not a total waste of time. Please do feel free to share your comments.

1) Source of Data

I used the official site Corona Virus Data for UK to pull the actual data. It has data for cases reported and also the total number of deaths. I picked up data only for “Latest Cases”. “Daily Lab Confirmed Cases” to be precise. The other dimension I have used is the Area Name so that I can analyse it by City and Regions.

Catch-up

I will start from the point where I have loaded the data in SAC and have run the Predictive Scenario. If you want to get familiar with it or do it yourself then please follow this excellent blog Hands-On Tutorial: Predictive Planning by Sarah Detzler .

Actual data Analysis

Let me give you an idea on the kind of data we are dealing with. The key dimensions are time, nation/region with “confirmed lab cases” as the Key Measure. I hope the graphs/info below will make it very clear. You would notice that we have data till 15th July initially and on purpose. btw today is 21st July 🙂 (15th July is our first load).

Figure 1: The image above shows data for England.

Figure 2:Graph to show the trend

2) Running Predictive Scenario in SAC

The actual data was thrown into the predictive scenario factory of SAC and it came back with very interesting results. I asked the predictive factory to do the predictions for the next 90 days. Here is a quick look at some of the best and worst cases.

Best Predictive Score – East Midlands (Region)

Seems like we are looking at no new cases from second week of August. Great news in the making.

Worst Case – Somerset (Upper Tier Local Authority)

Clearly in this scenario it was not able to predict to an acceptable accuracy.

England

After having a look at the best and the worst case, I was keen to know what the prediction score was for England. As you would expect, with these pandemics, there is a long nagging tail. The predictive algorithm was able to match with the actual lines and you can see the long tail after 15th July.

Signal Decomposition

SAC predictive was able to differentiate the trend, in Red line and fluctuations in purple. Pretty impressive.

3) Looking at Predictions with Actual Data

After my close look at the predictive score and signal analysis I asked the factory to apply it to my actual data model. I also loaded Actual data till 18th July so that I can see how the predictions have worked since my last load.

Here is a look at a graph for England. The gap certainly decreased on 18th July and the both Actual and Forecast lines are coming closer. The Forecast then continues to go in the wave fashion as discovered earlier.

Overall I found this very engaging and really a valuable exercise. The predictive from SAC works in so many ways. It makes it simple and it makes it very engaging for even people like us who are not Data Scientists.

Now for the most important question. When will this pandemic end in the UK? Here is the prediction. Please do remember we are not counting any additional external factors here.  Only time  can tell on its accuracy. I will leave you all with these interesting graphs below. Enjoy!

England – Goes till October

South West – Lucky us, it is finishing by end of July – Fingers Crossed!