For linear regression, seems like a new feature has to be a linear relation with the target variable.

But If you make the new feature for the Xgboost, what do you have to consider to make a new feature for a xgboost or lightgbm ?


1 Answer 1


There is nothing that needs to be preconsidered when a new feature is presented to LightBGM or XGBoost.

Even though the two mentioned methods work are nonlinear methods, there is no restriction on use them, the method itself will select or not the new features.


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.