imputation error
Closed this issue · 1 comments
For the returned OOB error, is there any general guidance on when there is too much error to likely be useful? Even if data- and context-dependent, I do not know how to interpret the NMRSE and the PFC, and I would grateful appreciate any resources to guide decisions with regard to these measures.
The best way to better understand the error measures is to run missForest in supervised mode, i.e. provide xtrue
and see what kind of errors appear. If your data contains missing values from the start, select a portion of it which is fully observed an generate some artificial missing values using prodNA()
and use this as xmis
while the fully observed part is used as xtrue
. However, as you say - the error rates are context dependent.