Regression trees

Hello community,

How are you? :slight_smile:

I am just writing to ask for the exit condition of a regression tree.

A decision tree used for classification could have the following exit conditions:

  • Entropy = 0
  • Threshold on depth
  • Threshold on information gain
  • Threshold on number of training examples

A regression tree uses variance reduction to decide what feature is the best for split and the entropy is not important in this type of trees so I was wondering what is its best exit condition. I think that a threshold on the returned variance reduction value could be the best (and also use a threshold on depth and number of training examples).

In advance, thank you very much for your time! :slight_smile:

I think all 4 of the conditions can be converted to the case when variance reduction is applied.

Variance = 0
Threshold on depth
Threshold on variance reduced
Threshold on number of training examples (in a leaf)

Cheers,
Raymond

1 Like