Due to recent updates, all users are required to create an Altair One account to login to the RapidMiner community. Click the Register button to create your account using the same email that you have previously used to login to the RapidMiner community. This will ensure that any previously created content will be synced to your Altair One account. Once you login, you will be asked to provide a username that identifies you to other Community users. Email us at Community with questions.
guidelines for base learner vs stack learner
Guru
Dear all, Im working on a classification problem with supervised learning, for which
I want to use the stacking operator to improve accuracy
1. while choosing stacking operator, im looking for Is there any guideline how to combine the operators.
Using Gradient boost alone gives accuracy 75.75%. I want to take it beyond 95% using stacking.
(ofcourse I'll look into the requirements of precision/recall).
I tried many stacking combinations, i couldnt make it beyond 73%.
2. Any theoretical/white paper reference / rapidminer resource/ book on how to choose the combination of base learners , stack learner for the maximum classification performance.
3. will it be useful to use non-neural net operator as base learner & neural net (deep learning ) as stacking learner?
i tried this combination, in my case, it reduces the performance to 55%.
thank you
thiru
I want to use the stacking operator to improve accuracy
1. while choosing stacking operator, im looking for Is there any guideline how to combine the operators.
Using Gradient boost alone gives accuracy 75.75%. I want to take it beyond 95% using stacking.
(ofcourse I'll look into the requirements of precision/recall).
I tried many stacking combinations, i couldnt make it beyond 73%.
2. Any theoretical/white paper reference / rapidminer resource/ book on how to choose the combination of base learners , stack learner for the maximum classification performance.
3. will it be useful to use non-neural net operator as base learner & neural net (deep learning ) as stacking learner?
i tried this combination, in my case, it reduces the performance to 55%.
thank you
thiru
0
Best Answer
-
Telcontar120
RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,635 
Unicorn
There is no way to know in advance (unless you are dealing with an artificial dataset used specifically for instructional purposes) whether an accuracy of 95% is feasible on your data (without overfitting).
Stacking is a kind of ensemble modeling. So is Gradient Boosting. They are different approaches but Gradient Boosting "alone" is still an ensemble method. In theory you may be able to improve your results by putting GBT into your ensemble as the base learner and then using something else in combination with it to help with those cases where it it not doing a good job, but again, it might not significantly improve the results on your validation set since further improvements may not be feasible due to overfitting which can come with having too complex of a model design. You might want to try using a DT as a stacker, this is often a popular choice.
7
