The simplest one you could try is majority voting (the predicted class would be the one that the majority of your classifiers are predicting). There are some easy variations as averaging predicted probabilities or a weighted average according to the performance of each single classifier.

Thank you Dr.Daniel

Hi,

In addition to the Vote approach, Stacking is another technique to go with. You could also apply Bagging and AdaBoost algorithms.

HTH.

Dr. Samer Sarsam

Hey there,

I think one of the most important issues is the combination stage (a.k.a fusion stage). To this end, you can use early or late fusion. In the early fusion, a combination of the content of each ensemble member, for example, its prediction score or its features are aggregated to learn a classifier (i.e., meta-learning). In the late fusion, the output of each ensemble member is combined by using techniques i.e., majority voting or weighted voting (as described in previous answers).

You can find more details about these techniques in our papers:

Hi,

I have used stacked generalization in classification tasks and gotten interesting results. However, the most important point in the use of Stacking is to apply class probabilities resulted from the 0-level classifiers instead of the single result for classes.

Hope this helps. If you need more information let me know to explain more.

Good luck,

Saeed