[SPARK-24710] Information Gain Ratio for decision trees

• Type: New Feature
• Status: Resolved
• Priority: Minor
• Resolution: Incomplete
• Affects Version/s: 2.3.1
• Component/s: ML
• Labels: bulk-closed features
• Assignee: Unassigned
• Reporter: Pablo J. Villacorta
• Created: 01/Jul/18 13:53
• Updated: 25/May/21 01:51
• Resolved: 25/May/21 01:41

Description

Spark currently uses Information Gain (IG) to decide the next feature to branch on when building a decision tree. In case of categorical features, IG is known to be biased towards features with a large number of categories. Information Gain Ratio solves this problem by dividing the IG by a number that characterizes the intrinsic information of a feature.

As far as I know, Spark has IG but not IGR. It would be nice to have the possibility to choose IGR instead of IG.