How to plot ROC curve for SVM classifier results?

8 次查看（过去 30 天）
Valeska Pearson
Valeska Pearson 2013-10-15
回答： Ilya 2013-10-16
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)

请先登录，再回答此问题。

回答（1 个）

Ilya
Ilya 2013-10-16

类别

在 Help Center 和 File Exchange 中查找有关 Statistics and Machine Learning Toolbox 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by