3

The textbook tells us that we should choose an attribute with the maximum information gain to split a decision tree. My question is what if all information gains are zero? Should we stop splitting or we split the tree with all attributes?

An example of this question is $Y=a~XOR~b$. To determine the value of $Y$, the information gains of $a$ and $b$ are zero. How do we build a decision tree for this question?

azure
  • 31
  • 1
  • 2
  • If the information gain is zero, there's no further purpose of splitting unless a combination of features yields information. I'm looking into that very question now: http://stats.stackexchange.com/questions/259176/can-decision-trees-look-multiple-levels-deep-when-selecting-features-to-maximize – Brian Bien Jan 31 '17 at 17:25

0 Answers0