Re: Request for Comments: New IPv6 OS detection machine learning engine

[Mathias Morbitzer <m.morbitzer () runbox com>]

On 02/21/2017 05:34 AM, Varunram Ganesh wrote:
> >
> >
> >     5) As we already thought, having 695 features is quite a lot.
> > Approaches to reduce the amount of features could be for example
> >
> >     using neural networks or principal component analysis (PCA). We
> >     did play around with such things a bit before, but it might be
> > interesting
> >
> >     to have another look.
> >
> >
> > I'm not exactly familiar with these either, but it definitely sounds
> > like it's worth a look!
>
> >Unfortunately, me neither. Anyone who is is welcome to apply to the next
> >Google Summer of Code!
>
> I think PCA might be a better choice considering we have a dataset of 
> slightly greater than 300 fingerprints and for neural networks to work 
> correctly, we would need as many fingerprints as features. That being 
> said, as you mentioned, it'd be helpful to have more fingerprints to 
> make the algorithm better (and maybe implement neural networks in the 
> future).
Agreed. We will hopefully have a closer look at PCA at this year's GSoC.

Last summer, we also tried to use the OpenCV function 
GetVarImportance(). [1] From its description, it should "return the 
variable importance vector, computed at the training stage", so we were 
hoping to get an indication on which features are most helpful for 
classification. However, the numbers we received didn't really help us - 
maybe we should have another look at this.

[1] 
http://docs.opencv.org/3.0.0/d0/d65/classcv_1_1ml_1_1RTrees.html#afc0c6f0c53e11d04ec1d1e176b3ba07f
_______________________________________________
Sent through the dev mailing list
https://nmap.org/mailman/listinfo/dev
Archived at http://seclists.org/nmap-dev/