Large historical data ≠ perfect predictor of future events

eNitiate | Historical data | City of Joburg | One Millionth Twitter Follower Tracker | 9-29 June 2020 | Banner

The era of widespread data-driven decisions has dawned on us, thanks to COVID-19.

This is the exciting time, says our founder – Bra Willy Seyama, a self-confessed data analytics fanatic.

But alas! Predictions of COVID-19 infection and death rates have been a challenge around the world, despite the increasing mountains of data.

Take the United States for instance, where COVID-19 death projections have varied between 100 000 and over 2 million

The White House predicted back in April this year that the US was going reach 74 000 deaths by August

As at 21 July, the tally was already 142 000 – almost double the prediction above.

We thought more historical data ➡ better predictive analytics?

It turns out historical events are not always a perfect benchmark for the prediction of future events, the latter which by their very nature are unknown.

This is especially so for a novel disease such as COVID-19, about which a lot is still being discovered to this day.

Add to the above the human behaviour aspect – typified by the lackluster adherence to measures for curbing the community spread of the virus, the metrics and algorithms used in the modeling of the data, and the genius involved in interpreting the results.

All the factors above play a significant role in the accuracy of the predictions.

 So why did we decide to develop an eNsight about this topic? Read on.

Table of Contents

Inspiration for this eNsight

We had a taste of inaccurate predictive analytics while we were tracking Johannesburg metropolitan municipality’s official Twitter account, as the city progressed towards earning its one millionth follower – a milestone that was reached on the 8th of July 2020.

The experience from the tracking referred to above inspired this eNsight.

Where the #OneMillionthTwitterFollower campaign started

While minding our own business on Twitter in the first week of June 2020, we noticed that @CityofJoburgZA – Johannesburg municipality’s official Twitter account – had followers in the 994 000 range. 

Out of curiosity, we asked two questions:

  1. When is the City of Gold going to clock its one millionth Twitter follower? 
  2. Which other African cities have achieved this milestone?

The two questions above led to the kicking off of the #OneMillionTwitterFollowers campaign, one of our many passion projects. 

We launched the campaign on the 8th of June with daily tracking of the municipality’s Twitter account’s new followers. 

Prediction of the #OneMillionTwitterFollowers date

Our initial prediction

In order to answer the first question, we needed to develop a predictive model for daily new Twitter followers for @CityofJoburgZA. 

For this purpose, we collected data that went back to the 8th of April 2020. 

This is what the trend of daily new followers for the 2 months to the 8th of June looked like:

eNitiate | Historical Data | City of Joburg | One Millionth Twitter Follower Tracker | 8 April - 8 June 2020

Our initial data modeling indicated on the 8th of June, and based on the collected historical data, that @CityofJoburgZA will earn the millionth follower by the 22nd of June 2020.

On adjustment of our data modeling, the predicted date moved to the 29th of June 2020.

It is on the basis of the adjusted prediction that we published our first tweet of the campaign:

In the short video below, Tumelo Komape – the Deputy Director of Online Media Communications at the  Johannesburg municipality – indicates in a Zoom chat with Bra Willy Seyama that, in his projection made in 2015, the 1 millionth Twitter follower milestone was set for achievement by June of 2019 already!   

Our revised predictions

As at the 20th of June, it became clear that our adjusted prediction of 1 million followers by the 29th was inaccurate. 

eNitiate | Historical Data | City of Joburg | One Millionth Twitter Follower Tracker | 9-29 June 2020

As the blue line graph above shows, @CityofjoburgZA was going to be 1 225 short of 1 million Twitter followers by the 29th of June.

Question was, what is going to be the new date for #OneMillionthTwitterFollower?

After another adjustment of our data model, we identified the new date as the 9th of July.

As it turned out, even the third predicted date – captured in the tweet above – was incorrect. 

But thankfully, it was close enough. The final predicted date was the 8th of July, which we were able to determine 3 days beforehand.

Feeling confident that our data modeling was finally precise, we could not wait for the eventual happening of the #OneMillionthTwitterFollower milestone.

We started tracking the progress more frequently from 7:25pm on the 7th of July.

And finally, the anticipated moment arrived!

You can find the rest of the #OneMillionthTwitterFollower tweet thread here.

Summary findings from the #OneMillionthTwitterFollower campaign predictive analytics?

There is a lot that goes into predicting future events, more than just the historical data. 

Novelty, human behaviour, metrics, data modeling and expertise also play a role.

In addition to the above, predictions become more accurate as the dates of the anticipated future events get closer.

Johannesburg municipality has the highest number of Twitter followers in Africa

In answering the second question found in the section about where the #OneMillionthTwitterFollower campaign started, we also found that @CityofJoburgZA is the official municipality Twitter account with the largest followers in Africa.

You can read more about the top 18 African cities on Twitter in the related eNsight below.

Did you find this eNsight valuable?

Sharing it on social media will ensure that many more people can benefit.

Select your preferred social network found on the right sidebar to share it.

Select your preferred social network found in the footer here below to share it.

It will take only seconds.

Related Posts:

Related Post:

the love.

share with the world

the love.

share with the world