I have three logistic regression models that were developed independently from each other on different datasets. However all of them describe the same phenomenon and predict the probability of occurrence of the same event (e.g. whether the client buy the product or not). The features in all models are also uncorrelated in the models and between them, so there is no situation that the variable from model 1 is correlated with variable from model 2. The outcomes of the models are also uncorrelated. I believe that the combination of these models will have positive effect on accuracy improvment and I would like to obtain one prediction for every client, also the probability. I cannot change anything in the models. What is the best way to combine predictions of the models?

One of the solution that comes to my mind is to develop new model, also logisitc regression, with predictions from the three models as explenatory variables. Any idea will be appreciate.

  • 2
    $\begingroup$ sounds like you want to combine the models via ensemble methods. Not sure if you use R, but either way, this is a good introduction cran.r-project.org/web/packages/caretEnsemble/vignettes/… $\endgroup$ – Jeff Mar 23 '15 at 14:51
  • $\begingroup$ Thanks, you helped a lot. Stacking is what I was looking for and it seems it is kind of ensemble learning. $\endgroup$ – hakubaa Mar 23 '15 at 15:10

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.