How to estimate AIC for SVM

Hi,

Is there a way to estimate AIC for an SVM model? Or is there any other type of measure that can be used to compare different SVM models.

Thak you

[Vidya Viswanathan]

Hi Chaturika,

I am aware that there is a function to estimate AIC (Akaike's Information Criterion) for an estimated model that is available with the System Identification Toolbox. However, from my understanding, that functionality seems to apply to systems and I am not sure if you can use it directly on an SVM classifier object. For your reference, the following is the documentation link for the function:

http://www.mathworks.com/help/ident/ref/aic.html

However, there seems to be an alternative function that is used specifically to compare the accuracies of two classification models. This function is called "compareHoldout" and the following is the documentation link for the function:

http://www.mathworks.com/help/stats/compactclassificationecoc.compareholdout.html

This could probably help you compare two different SVM models.

Regards,

Vidya