app roc curve different to perfcurve

Dear all,

I have a question, I want to overlay two roc curves, but my surprise is that when I use perfcurve the AUC is much higher that the one otained in the app. I think the one of the app (machine learning toolbox makes more sense) according to the accuracy.

How is possible to get a 5% of difference in AUC of the same classifier with the same data between the machine toolbox and the perfcurve.What is the difference between both implementations?

[~,score_svm] = resubPredict(classificationSVM);

[Xsvm,Ysvm,Tsvm,AUCsvm] = perfcurve(trainingData.Eval,score_svm(:,1),1);

Thank you

[Ive J]

Classification learner app calculates the AUC using the validation set, not the whole sample (perfcurve). If you set validation to Resubstitution validation in the learner app, the AUC from learner app and perfcurve should be same (this is only for comparison, and in this way your model is not proteceted against overfitting)

[Esmeralda Ruiz Pujadas]

Thank you, you are right. All clear. Thanks