What is Predicted R Square?💡

As I talked yesterday, R Square fails to show the complete picture of the model performance. You would never know if your model is overfitting.:thinking:

Predicted R Square tells you how well your model is predicting on new data points.

Simply put, if your model does well on training set and does bad on predicting for new data points, the Predictive R Square will be very low. Whereas, the R Square could still be extremely great.:dizzy_face:

But how is it calculated?

  1. A data point is removed or held out from the training set.
  2. Regression line is calculated.
  3. Line is evaluated on the held out data point.
  4. Repeated for all the data points.

The Predicted R Square will be distinctly small if the model is overfitting the training data.:bulb:

Therefore, it wouldn’t be wise to chase a high R Square value all the time.

Furthermore, it also cannot be said that a low R Square value is always bad.

It would depend on the domain you are working in, and what would be defined as a good enough R Square.:bulb:

#datascience #machinelearning