0
$\begingroup$

I'm making a predictive model. I'm thinking of using MI but not sure which imputation method to use. Is there some metrics or graphs one can compute on the data to see which method is best for which predictor? or maybe using many different methods and compare them? if so what should I watch out for when comparing them?

Lastly, how does MI work with k-fold cross validation? Say I want to do 50 of 10 fold cross validation with a data that is multiple imputed with m=5, I would do the imputation first and then do the cross validation on each of the 5 sets of data, and then pool them using mean?

$\endgroup$
1
$\begingroup$

For more details about the Inputation techniques in predictive models see this paper: Research paper, it presents an extensive experiments using the imputation techniques to deal with missing data in Software Effort Prediction Concerning the combination rule, I think that median is more robust than mean, because is not sensible to outliers.

$\endgroup$
1
  • $\begingroup$ Welcome to CV. In general, "link-only" answers do not do well on CV and risk deletion. You need to clarify and stipulate the information in the link relevant to answering the OPs question. $\endgroup$ – Mike Hunter May 21 '16 at 14:41

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.