Iterative Dichotomizer 3
- tries to create a shallow tree
- greedy
- recursive depth first
Algorithm
- choose best feature using Information Gain Ratio or some other metric, dont pick this feature again.
- partition dataset using selected feature (all possible branches)
- repeat until break condition
Break condition
- homogenous or pure set → mark as classfication
- no information gain if split
- run out of features to split on