How do you deal with missing values when scoring a model? Can I use multiple imputation when building the model?
$\begingroup$
$\endgroup$
4
-
1$\begingroup$ 1. Which model? 2. It's obviously a possibility, people do this all the time. So what's your question? Some specific things you're worried about? $\endgroup$– GijsCommented Apr 28, 2017 at 9:07
-
$\begingroup$ It's an elastic net model. The question is whether I can use a multiple imputation method when building the model? I'm new to scoring, so I'm not sure how your going to deal with the missing in the scoring process. No one in the department has used a MI in the model building before. $\endgroup$– HerzriesigCommented Apr 28, 2017 at 9:43
-
$\begingroup$ @Gijs is right: some elaboration is still required (for me at least): with what goal are you building a model? Do you mean to replace missing values in the scoring of variables when using a prediction model after its development, or do you mean to replace missing values in the model development dataset? These questions might have already been asked here. Additionally, what kind of model elastic net model are you fitting (i.e. what are your variables like)? $\endgroup$– IWSCommented Nov 6, 2017 at 11:57
-
$\begingroup$ @IWS the goal is to classify good and bad customers. I mean imputing during the scoring. I’m now aware that it is perfectly achievable to do so, it’s only quit a bit more work when the model is implemented. I’m fitting a logistic regression elastic net, not my 1st pick, but it works and keeps the non-tech dinosaurs in management happy $\endgroup$– HerzriesigCommented Nov 6, 2017 at 15:17
Add a comment
|