How to calculate scores from neural network classification ?

Question

Pradya Panyainkaew 2018-4-16

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/395138-how-to-calculate-scores-from-neural-network-classification

Hi everyone

I try to classify binary classes problem by using neural network classification (the raw data are attached in this post) and evaluate this network performance by using "perfcurve" function to find the area under curve (AUC) value.

The code below show how I classify my data.

 t_train=zeros(2,size(TrainSet_holiday,1));
t_test=zeros(2,size(TestSet_holiday,1));
for b1=1:size(TrainSet_holiday,1)
    if  Y_train_holiday(b1,1)==0
        t_train(1,b1)=1;
    else
        t_train(2,b1)=1;
    end
end
for b2=1:size(TestSet_holiday,1)
    if  Y_test_holiday(b2,1)==0
        t_test(1,b2)=1;
    else
        t_test(2,b2)=1;
    end
end
x_train=TrainSet_holiday';
x_test=TestSet_holiday';
setdemorandstream(391418381)
%%%%%%%%This code below is network training process %%%%%%%%%
net = patternnet(10);
view(net);
[net,tr] = train(net,x_train,t_train);
nntraintool
figure(1)
plotperform(tr)
testX = x_train(:,tr.testInd);
testT = t_train(:,tr.testInd);
testY = net(testX);
testIndices = vec2ind(testY);
figure(2)
plotconfusion(testT,testY)
[c_train,cm_train,ind_train,per_train] = confusion(testT,testY);
fprintf('Percentage Correct Classification of train set   : %f%%\n', 100*(1-c_train));
fprintf('Percentage Incorrect Classification of train set : %f%%\n', 100*c_train);
figure(3)
plotroc(testT,testY)
title({'ROC Curve of Train Set for Classification by Neural network classification',['Cluster number : ' num2str(ClusterNo_holiday) ' on holiday']})
%%%%%%%%This code below is predicted output by using trained network %%%%%%%%%
Output_predicted=net(x_test);
testIndices_output = vec2ind(Output_predicted);
figure(4)
plotconfusion(t_test,Output_predicted)
title({'Confusion Matrix of Test Set on holiday',['Cluster number : ' num2str(ClusterNo_holiday)]})
[c_test,cm_test,ind_test,per_test] = confusion(t_test,Output_predicted);
fprintf('Percentage Correct Classification of output from prediction in test set   : %f%%\n', 100*(1-c_test));
fprintf('Percentage Incorrect Classification of output from prediction in test set : %f%%\n', 100*c_test);
figure(5)
plotroc(t_test,Output_predicted)
title({'ROC Curve of Test Set for Classification by Neural network classification',['Cluster number : ' num2str(ClusterNo_holiday) ' on holiday']})

However, the "perfcurve" function require variable name "scores" and I don't know how to find this variable from neural network toolbox. Thus, are there any ways to solve my problem.

Thank you in advance for your suggestion

Pradya