I tried using MICE (Iterative Imputer from SKlearn) for imputing missing binary variables.
It works but imputed values are non-binary. it seems to be populating with probability (impressive that related columns add-up to 1. please see image of one such set of variables).
Now I need help to choose an option
- should i keep them as non-binary
- convert them to binary based on probability. i.e. in the example posted below should i convert imputed rows to non-smoker?
- not use MICE for binary features
Thanks in advance