Hi Maytham,

You could use either "Vote" or "Stacking" method.

HTH.

Samer

Hi Samer,

Could you please give me some more details about how these two methods work.

Thanks.

Vote provides a baseline method for combining classifers. The default approsch is to average their probability estimates.

However, for Stacking method, please have a look at this:

http://machine-learning.martinsewell.com/ensembles/stacking/

HTH.

Samer

Hi Maytham,

Some hybrid classifiers have been well-discussed in the attached papers. Hope this helps!

http://www.sciencedirect.com/science/article/pii/S0969698915300060

Thank you very much Samer.

Thank you very much Watheq.

first SVM to  eliminate  unwanted classes  and then KNN to form clusters.  in case  needed.

Thank you very much Antonios.

Thank you very much A. Padma.

Dear Maytham,

I do have an answer for the the question but it would finally conclude what Antonios Georgantzoglou is saying. So my support to that answer gives more support to you to implement. 

Why not try locally (linear and nonlinear) SVM (LL SVM) classifiers. In short, it designs SVM for some local neighborhood of k-data. Alternatively try ALH approach, but LL SVM is more accurate, although slower. Download my papers on ALH and LL SVM from research gate site here. Good luck.

Thank you Vojislav.

You can do  it by stacking the classifiers or get the result from clssifiers separately and then do some sort of voting to get the results

Thank you Akhtar.

If your dataset is known, fixed and relatively reduced in size, k-NN would fit much better. Although SVN can provide better accuracy in some classification problems, I would not recommend it due to several drawbacks: lots of parameters to be tuned, increased complexity, difficulties to be extended for multi-class classification.

Hi Ioan Sorin Comsa

What do you think would be the better replacement of SVM in any hybridization approach?

When dataset is not fully labelled.

Thanks

To Maitreyee Dey,

If your dataset is not fully labeled, then you should do it somehow especially if you are dealing with real-time classification processes. You can use some hybrid k-means clustering approaches to label the data and find the patterns. Then you can use some types of neural networks to sort out with approximation and generalization issues.

Thanks Ioan Sorin Comsa ,

I did that earlier but althoug trying to do some semi-supervised approach.

Any how thank you for replying my query.