Atlas record: https://www.aitimeline.dev/en/events/evt-0151/
Research · March 1986
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.
Event record
- Event date
- March 1986
- Timeline date
- Event date
- Verification
- Sources gathered automatically · September 17, 2026
- Lines
- ID
- evt-0151
Machine Learning volume 1, issue 1, 1986.
Related events
Earlier
Later
- Related AdaBoost 1995links to this record
In practice the method was most often applied to decision trees.
- Related Weka 1999links to this record
Decision trees and other classical methods became available as ready implementations.
- Builds on Random forests 2001links to this record
A random forest is an ensemble of decision trees voting for a class; the tree as a classifier is the subject of the 1986 record.
- Related Gradient boosting 2001links to this record
The special version of the method, TreeBoost, is built on regression trees: the components of the additive model are trees.