In ensemble learning, model stacking is a good way to improve the performance of a single model. However, models chosen to be stacked must have least correlation in order to exploit the performance of stacking.

So my question is how can I check the correlation of models in stacking?

  • $\begingroup$ Run a validation set through the models and check the correlations of their outputs. $\endgroup$ – Dan Dec 8 '17 at 10:24
  • $\begingroup$ @Dan, can you please provide more detailed process for it? $\endgroup$ – GoingMyWay Dec 8 '17 at 12:58
  • $\begingroup$ The process I'm suggesting is very simple. I'm assuming you have split your data into training, validation and test sets. You fit you different models on the training set and then score the models on the validation set. So now for each model you have a vector of scores (e.g. probabilities) which you would then feed into another model if you're stacking. If you want to know the correlation of the models, before you stack them, simply compute the correlation matrix on those vectors of output scores you created by scoring your validation set on each of your models. $\endgroup$ – Dan Dec 8 '17 at 14:22

Your Answer

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

Browse other questions tagged or ask your own question.