Treatment learner

From [1]: In Data mining, a treatment learner finds rules that change the expected class distribution (compared to some baseline).

Classifiers are used for recognition tasks; e.g. find what toys to remove from the assembly line because they are defective.

Treatment learners are used for planning some minimal action to improve the odds that something will be later be recognized as belonging to some class. A treatment learner could be used to make repairs to the defective toys rejected from the assembly line.

Treatment learners are all about minimality - what is the least you need to do to most effect something.