2
$\begingroup$

I'm working with several estimators of all kind. Then, I want to stack these estimators, and the best is if they have low correlation between them.

I suppose that the correlation method depends on the type of dependent variable, if it's categorical or numerical. In my case, it's categorical, and the estimators are classifiers.

How can I do the correlation between two estimators?

$\endgroup$
1
$\begingroup$

There are probably many options, I'd look at inter-annotator agreement.

An alternative option would be to to evaluate one (e.g. by accuracy or f1-score) using the other as the gold standard.

$\endgroup$
0
$\begingroup$

If I am not wrong this already exists : Random Forest ... Basically, a lower correlation among ensemble model weak classifiers increases the error-correcting ability of the model. Therefore, it is preferred to use models with low correlations while creating ensemble classifier. Random Forest uses some clever tricks to make its decision trees likely to be independent.

$\endgroup$

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.