One Big Trouble with Predictions
By Soumyadip Pal
Predictive Analytics is magical. Imagine a computer being able to tell you exactly what is going to happen. Fantastic, isn’t it? What can be a problem with it?
Casey “The Old Professor” Stengel had once said: “Never make predictions, especially about the future.” Yet, that forbidden territory is exactly where predictive analytics dares to venture.
The result, as you may imagine is always wrong. As the saying goes, there are two types of forecasts - lucky and wrong. So, when predictive analytics is not wrong, it just is plain lucky.
What really is the problem with predictive analytics?
Many things in business are not exactly a rigid science. Our expectations of the accuracy of predictions must then be tempered by the historical data that goes into training the model. Naturally, the model cannot account for new events that hadn’t occurred in the past and were not recorded in the training data.
One may say that the solution is to take the unknowns into account. Is that even possible? How do you account for something that you don’t know, and hence, cannot anticipate?
The unknowns that can be factored in a model would be the “known” and “anticipated” unknowns. Reality, though, is filled with “black swans”, that is, things that are outside the “known” unknowns, and thus are tremendously difficult to predict.
Predictive models must be made to account for truly unknown unknowns, and that is way more difficult than it sounds. This also means that almost no model takes black swans into account.
Are predictive analytics junk? No! They certainly have their place in business. They’re extremely helpful as long as we do not expect pinpointed accuracy from the models. Tempering down the expectations is all the more crucial when we are modelling for high-stakes battles. The urge is to build complex models and then the errors and inaccuracies balloon up. Here we go again…
>> To get detailed information on Business Analytics with R, or if you are looking for a certification in Business Analytics please go to Manipal ProLearn.
Soumyadip Pal is a retail analytics professional and a passionate educator with more than 8 years in the industry and more than 7 years in the academia, currently working as a consultant with Manipal Prolearn.