There are people who are using Hierarchical SVM for multi-class classification. But I think it simplifies the original problem and doesn't serve the actual purpose. OnevsAll though similar to this approach but still uses all classes at each classification. Any idea?
Dear Amir,
you there are multiple ways to combine classifiers. In the standard way SVM is a binary classifier and if you want to work with mutli-class classification the library LibSVM does binary models of every group against every other group. Finally, the predictions are combined by majority vote. That is hidden, you don't see that.
You could also do that hierarchically have an SVM at every knot and while predicting you follow the hierarchy. That is a different combination scheme. (Numerous these schemes exist)
Best, Thomas
Dear Amir Khan
This paper might be helpful (https://ieeexplore.ieee.org/abstract/document/7280489/)
Kind Regards
Qamar Ul Islam
Dear Amir Khan
I *think* this is where it started:
https://core.ac.uk/download/pdf/71880764.pdf
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking