> A basic k-means or naive Bayes approach will show up in ML textbooks, but those aren't clearly different from "use some statistics to make a prediction".

Ha, sounds like a classification problem! Let's use ML to find the boundary.