0
$\begingroup$

I am currently constructing a model that uses last year's departmental information to predict employee churn for the current year. I have 55 features and 318 departments in my data set. A good portion of my independent variables are correlated, and because of this, I believe that performing a ridge regression on my data will lead to optimal predictions when I bring the model into production. I have studied ridge regression and understand that the lambda coefficient computed for a given predictor can minimize the effect that predictor has on the model to next to nothing. Does this mean that performing a ridge regression means I don't have to bother with variable selection? If I do need to perform a variable selection technique, would implementing a stepwise regression and then using those selected variables in my ridge regression be a valid approach at variable selection? I already posted this question on stack exchange but was informed that stack exchange was the better platform to ask statistical questions. I am sorry for the confusion.

$\endgroup$
3
$\begingroup$

Ridge regression is a good choice. But it is necessary to run it on all candidate variables. Doing variable selection before ridge regression will invalidate it.

The beauty of ridge regression is that it doesn't spend any information in the data trying to select variables. Methods that select variables do not result in stable sets of 'selected' variables. More details are in my RMS course notes.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.