I tried to shrink all features that create multicollinearity problem in my model. For this I use VIF for understanding what level of $\alpha$ coefficient for lasso will be enough (calculate VIF for features where coefficient not equal to zero and waiting when all VIF will be less than 5). Does it make sense for you or it is strange idea.?

  • 1
    $\begingroup$ Why are you shrinking variables involved in colinearity? $\endgroup$ – Peter Flom May 14 '18 at 22:40
  • $\begingroup$ Why would VIF depend on $\alpha$? $\endgroup$ – The Laconic May 15 '18 at 0:35
  • $\begingroup$ The main question for my model is to rank feachers. So model should be interpretable and there should be no multicollinearity problem because it have influenced in values of coefficient. $\endgroup$ – Anna Bartunova May 15 '18 at 1:58
  • $\begingroup$ Why I connect VIF and alfa.. it was idea (maybe not correct) that with Alfa level that decrease for example x3,x5 to zero VIF calculated on all feachers except x3 and x5 will show as that there is no mo multicollinearity in our model $\endgroup$ – Anna Bartunova May 15 '18 at 2:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.