Suppose I have a linear model predicting class-membership from a set of predictors. Now, I am going to classify a new observation which has, however, some predictor values missing. How can I deal with such situation? I know there are methods for imputing the missing values but I would like to avoid this and to use only measurements that were really made.
One possible way to deal with this situation without imputation would be to refit the model without the missing predictors.