Does encoding missing data with fixed values help in classification?

Question

I have a lot of missing values for some variables in my data (70-80%). I have seen some people deal with missing values this way: encode the variable with missing values as 0 or 1. Where 0 is the value is missing and 1 as non missing.

I want to know if that technique is of any use, because I don't see any valuable information algorithm would able to pick from such variables. Also I am thinking of imputing them using mice but the problem is that in future use, we may not be able to get those variables with missing data, so the train and test set will have different number of columns

Answer 1

You can treat the mere presence of any value as signal - hence the 0 or 1.

What help this would be to your project depends on your dataset.