Induction of Decision Trees
Quinlan described ID3: the algorithm builds a tree, each time choosing the feature that most reduces uncertainty.
Why it matters
Learning produced a result a person can read and check, a rarity that made trees the standard in applied work.
The feature is chosen by information gain, the same notion of entropy Shannon defined. A tree translates into a set of if-then rules that can be shown to an expert. ID3 became C4.5, one of the most used algorithms of the 1990s, and random forests and gradient boosting grew from trees later.