End of statistical models?

With greater computational powers and ever increasing easier methods of gathering and storing data, machine learning methods are fast catching up.

But does it mean that statistical models are now a thing of the past? The answer to this question boils down to the objective that we want to achieve using the data at hand.

If interpret-ability of the results are more important than the predictions, we turn to our classical & traditional statistical methods. In simple words, if we want to examine the relationship between a variable and the outcome, statistical modelling is the clear winner.

For e.g. Let's say we have a sample of house prices data in an area and we want to find out if there is a relationship between the house prices and the "Age" of the house, i.e. how old the house is. The same can be deduced using statistical modelling techniques.

However, if predictability is what we are aiming for, then machine learning methods are the way forward. They are very good at identifying complex non linear patterns in the data and can give accurate predictions.

The above can be clearly summarized using the snippet below:

So are we saying that statistical models can't be good at predictions?

The answer is NO.

We expect the statistical models to be at par or outperform the predictions from machine learning methods when most of the assumptions that we make about the data before performing statistical modelling are satisfied.

However, if our focus is not much on the predictions, statistical methods can still be very useful in predicting significant relationships between the variable and the outcome.

So its not the end of statistical models after all. The classical methods still continue to play a major role in solving business problems using data science.

#datascience #machinelearning #analytics #statistics #data #modelling #deeplearning #neuralnetworks #business