I have a training dataset of 2000 instances with 400 features. These features are values of histogram bins. Some of these features have constant and zero values for all data instances, hence scaling can not be applied while using svm() and tune() functions. If they are removed, i face a problem that testing dataset do not have those features with zero or constant values. And i cannot create a model with partial features and then test it with complete set of features. Ay help in this regard is highly appreciable.