Inputs for the ROC curve

Question

I have a 2 column matrix, where in each row are observations for healthy (column 1) and not healthy (2 column) patients. Also, I have 5 partition values which should be used to plot ROC curve. Could you please help me to understand how to get the inputs from this data for the perfcurve function?

Thank you for any reply!

Answer 1

I've made a small script that shows the basics of a perfcurve given a two column matrix input. If you execute this in MATLAB and take a careful look at it then you should have no trouble using perfcurve

%Simulate your data as Gaussian data with 1000 measurements in each group. 
%Lets give them a mean difference of 1 and a standard deviation of 1.
Data = zeros(1000,2);
Data(:,1) = normrnd(0,1,1000,1);
Data(:,2) = normrnd(1,1,1000,1);

%Now the data is reshaped to a vector (required for perfcurve) and  I create the labels.
Data = reshape(Data,2000,1);
Labels = zeros(size(Data,1),1);
Labels(end/2+1:end) = 1; 
%Your bottom half of the data (initially second column) is now group 1, the
%top half is group 0. 
%Lets set the positive class to group 1.
PosClass = 1;

%Now we have all required variables to call perfcurve. We will give
%perfcurve the 'Xvals' input to define the values at which the ROC curve is
%calculated. This parameter can be left out to let matlab calculate the
%curve at all values. 
[X Y] = perfcurve(Labels,Data,PosClass, 'Xvals', 0:0.25:1);

%Lets plot this
plot(X,Y)

%One limitation in scripting it like this is that you must have equal group
%sizes for healthy and sick. If you reshape your Data matrix to a vector
%and keep a seperate labels vector then you can also handle groups of
%different sizes.