1

I want to use elastic net (lasso + ridge) method for feature selection process. I can't understand why does the ridge method gives me the grouping effect for correlated variables. Can anyone explain that please?

Corel
  • 737
  • 7
  • 18
  • 1
    @kjetil Thank you I think that the explanation in that post sums it up pretty nicely. – Corel Dec 10 '17 at 17:14

1 Answers1

0

The L1 norm tends to overfit the most contributory $X_i$ variable. With ridge regression a bias is added that can reduce the propagated error of a parameter of interest, for example, see this. Alternatively, ridge regression can be used to reduce covariance if one uses a smoothing factor appropriate to that task.

Carl
  • 11,532
  • 7
  • 45
  • 102