How to train an SVM classifier and calculate performance

Hi all,

I was already browsing through some similar question, but I still don't understand completely how to train an SVM classifier with matlab and afterwards calculate performance measures like AUC, Accuracy asf.

I managed to use fitcsvm to train a classifier and using leave-one-out cross-validation:

model=fitcsvm(data,groups,'Standardize',true,'ClassNames',{'group1','group2'},'Leaveout','on')

This works well, but how to calculate performance measures of my classifier after this step and plot the results?

Answer 1

[Swarooph]

You could do one of several things:

1. Resubstitution Loss calculation using resubLoss function

2. Loss calculation using loss function

3. Performance evaluation using perfcurve -- (Another link - Evaluate Classifier Performance Using perfcurve)

[holistic]

Thank you! Perfcurve is what I want, but I'm not sure how to initialize it with the output from fitcsvm after leave-one-out cross-validation, i.e. where to find the "scores"?

[Swarooph]

If you look at the examples in the documentation, it seems to be using fitPosterior followed by resubPredict function.

Answer 2

[Samaneh Nemati]

you need to pass the output of svm classification (model) to predict function to get "label" and "scores".