How can I find the different score and threshold of a treebagger for ROC plotting?

Question

May 2018-11-19

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/430809-how-can-i-find-the-different-score-and-threshold-of-a-treebagger-for-roc-plotting

编辑： May 2018-11-19

Data.mat

I am trying to plot a ROC to know the preformance of tree bagger. I use the attached dataset and the following code to perform the classifiaction. As my problem is a binary classification, I have two columns of score values from treebagger. My class of interest is '1' and I want to calculate the thresholds. My ROC curve is something wrong. Please kindly help me to solve it. Thank you.

load ('Data.mat');
Allfeature = D(:,1:14);
Label = D(:,15);
% Half for training
Trainfeatures  = Allfeature(1:544,:); 
Trainlabels =Label(1:544,:);
% Half for testing
Testfeatures = Allfeature(545:end,:);
Testlabels = Label(545:end,:); 
% tree bagger for classification
template = templateTree(...
    'MaxNumSplits', 2000);
TreeBagger = fitcensemble(...
   Trainfeatures, ...
   Trainlabels, ...
    'Method', 'Bag', ...
    'NumLearningCycles',100, ...
    'Learners', template,'ClassNames', [1; 0]);
[labels,score] = predict(TreeBagger,Testfeatures);
diffscore = score(:,1) - max(score(:,2));
[X,Y,T,~,OPTROCPT,suby,subnames] = perfcurve(Testlabels,diffscore,'1');
% plotconfusion(ORFTestlabels,ORFlabels)
[c,cm,ind,per] = confusion(Testlabels',labels')
plot(X,Y,'--','LineWidth',1)