Engineering missing values (NA) in categorical variables - Random sample imputation