Hi all,
I would like to know if it is possible to select important predictors out of all available list of predictors in case of MSN in yaImpute package in R. I went through some articles where they are using random forest for variable selection. I just want to use MSN in my case. I am predicting some forest structure attributes based on airborne laser scanner.
regards
Hari
Hi Hari
Yes you can. You first run varSelection from the yaImpute package. If you choose addVars as the method, then this finds the variables that strengthen the nearest neighbor imputation model; if you choose delVars then this deletes the variables that don't strengthen then model. You can plot the reults to see how mean distance decreases as variables are added.
The final step is to run bestVars on the output of varSelection. This ranks the variables in terms of importance for the imputation model (in technical terms, ranks them by generalized root mean square distance)
The important thing is that you don't have to use random forests to build your yai imputation model. If you select model=msn then most similar neighbor is used. [You could use random forests by selecting method=randomForest, but this is what you seem to want to avoid]
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training