I am currious how good such a procedure could be : I get some predictions of some 10 learners trained on train set and predicted on train set also . Then I am column binding those predictions to the original train. Could this be a valid procedure on improving learning process?


1 Answer 1


This procedure exists and is called stacked generalization or simply stacking. See stacking section from wikipedia page. Strating from there you can red more by following references from the page. The first paper on subject was published by Wolpert in 1992.

[Later edit]

Do not combine the results with original features, but keep only the predictions combined only with the target.

  • $\begingroup$ Tryied stacking before , we talked on a previous thread about this, hence it is not always improving , I arrived at this question $\endgroup$
    – gm1
    Commented Jan 11, 2016 at 11:52

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.