I have been reading the original article about AdaBoost and by comparing with other reading material it has come some doubts about this model. Please feel free to correct me if in any part I am wrong. For what I see this model uses something called a Decision stump which is only a node that considers. For example, in the following image:

enter image description here

Only one feature at a time is used for making the decision stump, am I right?

Also, and I was not able to find this part in the original algorithm, is it necessary to have a splitting criteria like Gini or Entropy? or in this case it only relies in the error that each weak classifier has? I mention this because I was not able to find some explanation about if its needed and when should I consider this splitting criteria.

Thanks, and sorry for this simple question.

  • $\begingroup$ Does this answer your questions: How to use decision stump as weak learner in Adaboost? $\endgroup$
    – chl
    Nov 22, 2020 at 19:26
  • $\begingroup$ Thank you @chl, but it answers it partially; because it does not mention in which part or if its needed to use Gini or entropy for splitting the decision stumps. $\endgroup$
    – Layla
    Nov 23, 2020 at 19:27

1 Answer 1


Only one feature at a time is used for making the decision stump, am I right?

In the sense that the trained decision stump only uses one feature, yes. But training the decision stump is like any other tree model: it chooses which feature to split on, and what point at which to split that feature, based on some impurity criterion.


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.