How do you handle and mitigate the effects of concept drift in machine learning models?

How do you handle and mitigate the effects of concept drift in machine learning models?