R-Squared is without doubt one of the hottest metrics to judge regression fashions. It’s taught in any statistics class and it’s one of many metrics applied in Scikit-learn.
Nonetheless, some doubts have been raised concerning the reliability of this metric. Within the notes for his course at Carnegie Mellon University, Professor Cosma Shalizi claims that R-Squared is ineffective.
So, ought to we utterly dismiss R-Squared?
I don’t suppose so.
I admit that this metric has one main flaw, however I additionally suppose we shouldn’t lose sight of the positives. On this article, I’ll clarify what’s flawed with R-Squared, and counsel a modification that makes it totally dependable.
To know what’s the downside with R-Squared, we first want to grasp its which means. And I imply the deeper which means, not the sloppy definitions that may be present in most sources.
Let’s begin with an instance. Suppose we’ve got a predictive mannequin (“mannequin A”) designed to forecast the promoting worth of a home.
Think about that our check set consists of 4 homes. We are able to visually test the…