Good point. But in production almost all prefer to...

# datascience

[khud]

Good point. But in production almost all prefer to have a simple model which can be interpreted well. Therefore the best way to deal with the issues is to train the simple model again (with new data) if it does not behave as good as it should.