1
$\begingroup$

When using rpart to create classification tree, the values for the relative importance of each predictor show up along these lines:

Var1: 33
Var2: 31
Var3: 25
Var4: 3

In my case Var3 is plotted as the root node. I expected that Var1 would have been the root node, given that it has the highest relative importance. Based on this, would it be reasonable to expect that Var1-3 would show up more and/or higher up towards the root of the tree? That question also applies to decision trees in general.

Thanks

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

Variable 3 is the predictor that provides the most separation in the two nodes after a single binary split. Predictors might show up multiple times further down the tree resulting greater overall importance.

Improve this answer
$\endgroup$
1
  • $\begingroup$ That makes sense. So in this case would I be correct in saying that Var3 results in the greatest degree of purity between each side of the split compared to the other nodes? $\endgroup$
    – The WhetherMan
    Commented Mar 3, 2016 at 22:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.