if seems that the domain of your training data for the decision tree does not match the domain of the input data for the decision tree predictor. For a data row with "new" values it is possible that the decision tree predictor cannot perform a prediction and hence returns a missing value as prediction which cannot be handled by the feature elimination node.
The easiest workaround for you is to append a "Missing Value" node to the predictor and remove the rows with missing values.