R classifier that handles NA's at prediction time Are there any high performing classifiers implemented in R capable of handling NA's at prediction time? By high performing I mean models that in general produce good results (random forest, SVM, Boosting, etc.). Currently, I was trying with randomForest package but when trying to predict an observation having in some of its columns NA's then, the result is just NA.
At training time I have no problem since I can guarantee that the training data will not have any NA's. For my use case imputation is not an option, so I am looking for a classifier that just can handle NA values (maybe by ignoring them) at prediction time. Note: my predictor variables are numeric and categorical.
 A: Gradient boosted trees handle missings. 
The popular implementations in R, xgboost and gbm, both create a missing node alongside the left and right nodes for each split. If there are no missing elements in the training data then the missing node prediction is the same as the root prediction$^1$. Otherwise the prediction for the missing node is learned.
Be very careful though, as it sounds like your training data and implementation data are very different. If the model expects to always be able to see a certain powerful predictor, don't expect good performance when that predictor is later masked. Consider if there is a way to make your training data look more like the data on which you want to implement your model. Boosted trees' handling of missing is a neat convenience when your training and scoring data both have equally distributed missings, but it is not a catch-all for what you are trying to do.
$^1$ Edit: apparently xgboost always sends missings to the right node if it does not encounter any in training, and fits them to the left or right node otherwise. It does not spawn third node.
A: I have done some comparison with the other methods and XGboost handle missing data better. data are not presented in the feature matrix treated as "missing" and for continuous feature, it learns the best direction to go when the value is missing. 
You can read more here 
