What is the difference between both of the methods? Is missForest a multiple imputation method? If so, how does it differ from MICE using RF?
Thank you
1 Answers
Well, the similarity is, both methods are employing random forests to get their imputations. How the random forests are employed for this task is different between these two packages.
- mice: gives multiple imputations
- missForest: only provides single imputation
But the difference is not only in single vs. multiple imputation, the difference is also about the whole process how the random forests are employed to get the imputed values.
Explaining the details would take way too much here. If you are interested to take a deeper look into the respective algorithms of the papers:
Also keep in mind, that there is mice.impute.rf and the CALIBERrfimpute package, which both implement random forest multiple imputation for mice. But these are only slightly different.
There is also an interesting study comparing missForest, mice.impute.rf and CALIBERrfimpute and non-random forest mice methods. (I guess reading this will make things clearer for you)
Related