When features have missing values, they can be filled in using the Missing Information Principle:
Fill in the data using the available information.
Methods exist such as Gibbs Sampling, but they are timeconsuming.