2
$\begingroup$

I am running a Lasso regression for a model with one target and several predictors. I have standardized the predictors (but not the target) before running the regression. The results I am getting confirm my notion of the phenomenon, and Lasso is doing a good job of reducing the coefficients of predictors which are not (or should not be) important to $0$.

My question is: How do I use the Lasso coefficients for predictions and reporting? Since I had standardized the predictor values, should I re-scale the non-zero coefficients before making predictions and reporting the regression equation?

P.S. I am running the model on data from a phenomenon I know well to get a good understanding of 'shrinkage' regression models.

Improve this question
$\endgroup$
1
  • $\begingroup$ do you use R or python libraries for the lasso regression? you might want to check if the correspondng function scales your variables by default. $\endgroup$
    – Edgar
    Commented Dec 18, 2019 at 17:26

1 Answer 1

Reset to default
4
$\begingroup$

It is a good idea to rescale the coefficients back so that a user can

  • interpret them in the perspective of &
  • use them for prediction with

the regressors on the original scale (as opposed to scaled ones).

In addition, you may or may not report the coefficients before rescaling. They give an idea how much the dependent variable varies with the relative variation in the regressors (relative to their scale).

Improve this answer
$\endgroup$

Your Answer

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

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