For numerical value, the median is used for imputing the data. If you go to a trained model blueprint and click on the missing value imputed, you will see how we perform the impute. We just impute using the medial, but if you click on copy and edit, you will be able to choose random imputing.
Now, for categorical features. We add a missing class to represent the missing values.
By the way, if you click on Copy and Edit (orange top right corner), you can not only edit the blueprint but see all the additional pre-processing
functionality we offer. Just click on a node and then the + sign to see the list of pre-processing approaches we have.