0
$\begingroup$

When we use SVM to do classification, we usually will use this kind of code:

[predict_label,accuracy,decision]=svmpredict(testlabel,testdata,model);

We all know the classifier was using probability to make sure if the predict_label is positive or negative, which is according to thresholds. And with each threshold, we can get a group of sensitivity and specificity, thus we can draw an ROC curve.

My question is the accuracy from the output, we always get a specific number, which is also called ACC. So which threshold is this specific number based on?

$\endgroup$
1
$\begingroup$

The reported accuracy is for the default threshold of $0.0$.

$\endgroup$
2
  • $\begingroup$ you mean when testdata is labeled with -1 and 1. $\endgroup$ Dec 26 '16 at 4:03
  • $\begingroup$ If you are doing two-class classification, LIBSVM will default to renaming your classes 1 and -1 regardless of what they were. Also, the decision values are not bounded by the class labels: they go beyond 1 and -1. $\endgroup$ Dec 26 '16 at 4:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.