How to plot ROC curve for SVM classifier results?

Hello experts,

I need urgent help please. I have training data en test data for my retinal images. I have my SVM implemented. But when I want to obtain a ROC curve for 10-fold cross validation or make a 80% train and 20% train experiment I can't find the answer to have multiple points to plot. I understand that sensitivity vs 1-specificity is plotted, but after svm obtain predicted values, you have only one sensitivity and one specificity. So I tried rocplot and the perfcurve, but I haven't got the ROC curve as would expected. It is frustrating because, if I give perfcurve the inputs like this X,Y,T,AUC]=perfcurve(testLabel,pred ,1); the testlabel is for one dataset, this only plot one (sensitivity versus 1-specificity) point, where is the round or stair roc curve values generated from. I just want a valid ROC curve code that works??

Here is my work, where I tried 10-fold cross validation:

labels=[1;1;1;1;1;1;1;1;1;1;1;1;1;1;1;1;1;... 1;1;1;1;1;0;0;0;1;0;1;0;0;1;0;... 0;0;0;1;0;0;0;0;0;0;1;0;0;... 0;0;0;1;1;1;1;0;0;1;1;0;0;0;0;0;0;... 0;0;0;1;1;0];

My two features I use:

TrainVec=[amountExudates ,GS];

k=10;

cvFolds = crossvalind('Kfold',labels, k);

cp = classperf(labels);

for i = 1:10

    testIdx = (cvFolds == i);   

    trainIdx = ~testIdx;   

    svm = svmtrain(TrainVec(trainIdx,:),labels(trainIdx),... 

        'Autoscale',true, 'Showplot',true, 'Method','QP',... 
                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          lot',false, 'Method','QP', ...     

        'BoxConstraint',2e-1, 'kernel_function','rbf','rbf_sigma',0.1);  

    pred = svmclassify(svm, TrainVec(testIdx,:),'Showplot',true);

    cp2 = classperf(cp, pred, testIdx);

    testLabel=labels(testIdx);

Then I tried

[tpr,fpr,thresholds] = roc(testLabel,pred);

 plotroc(testLabel,pred);

and I tried

% Xnew=TrainVec(trainIdx);

% shift = svm.ScaleData.shift;

% scale = svm.ScaleData.scaleFactor;

% Xnew = bsxfun(@plus,Xnew,shift);

% Xnew = bsxfun(@times,Xnew,scale);

% sv = svm.SupportVectors;

% alphaHat = svm.Alpha;

% bias = svm.Bias;

% kfun = svm.KernelFunction;

% kfunargs = svm.KernelFunctionArgs;

% f = kfun(sv,Xnew,kfunargs{:})'*alphaHat(:) + bias;

% f = -f;

 [X,Y,T,AUC]=perfcurve(testLabel,pred ,1);

figure;plot(X,Y)

Answer 1

[Ilya]

See this thread.