Can we use R2-Sqaure to validate our model?

Question

Can we use R2-Sqaure to validate our model? When does the R2-score metric make sense?

in progress 0
demogorgon 55 years 1 Answer 654 views Member 0

Answer ( 1 )

  1. R2 squared can be explained as [1-(unexplained variation-RSS/total variation-TSS)] which can also be written as MSS/TSS, only when the decomposition TSS=MSS+RSS is valid. And that is true in linear models only(can be proved by simple mathematics and a bit of algebra). As written above RSS explains the amount of unexplained variance of our model only in the case of linear models and the same is the case with MSS being explained variance. So R2-square for other families of models may be low and can still be a good fit and vice-versa. So no conclusion can be made through R2-square as it can only explain linearity.

Leave an answer

Browse
Browse