1
$\begingroup$

I am developing a prognostic index using the LASSO technique and wondering how to deal with the highly correlated predictor variables. Should I choose the ones I want to include in the LASSO a priori or is the LASSO programmed to select the best/better of highly correlated predictors variables? thanks.

$\endgroup$
1
  • 1
    $\begingroup$ That is one of LASSO inherit weakness. You could probably use elastic net to make use of a ridge penalty to make feature slightly "more uncorrelated". This post on: How does LASSO select among collinear predictors? is rather enlightening too, I think. $\endgroup$
    – usεr11852
    Jul 26, 2016 at 9:19

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.