2
$\begingroup$

I am using the Orange data mining tool to build and analyze models (decision tree, ANN, ...) predicting customer churn. As this is an imbalanced class problem (10% churn, 90% not churn), I need to oversample within the cross validation. However, I am not totally able to implement this by myself. Is there anyone with some Orange knowledge that could help me?

Thank you!

$\endgroup$
0
$\begingroup$

Orange does not have over/undersampling. Our reasoning is that if you model a problem with 10% positive class, than you should not train the model with 50:50 class distribution - it will not reflect the real life. However, there's an option in Orange in LogReg and Random Forest to balance class distribution, which considers class distribution when building a model.

$\endgroup$
  • $\begingroup$ Thank you for your answer. I was wondering where I could find this Option in LogReg and Random Forests? Do you have to activate this option or is it build-in? Because I don't see it in the API of Logistic regression or Random Forests? $\endgroup$ – Emma Bartholomeeusen Dec 15 '20 at 14:20
  • $\begingroup$ It is the class_weight parameter. $\endgroup$ – vijolica Dec 16 '20 at 8:12
  • $\begingroup$ Thank you. Are you somewhat familiar with Orange? Because I am struggling with a few basic operations on a datatable and looking for someone who could help me get started.. $\endgroup$ – Emma Bartholomeeusen Dec 16 '20 at 10:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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