have I calculated 'scores' correctly for ROC curve input from PLS-DA

HI

i've used plsregress() on some FTIR chemical spectrum data, to get a model which predicts a categorical variable (but because this is a regression, i've set these to '0' and '1')

I want to test the model using an ROC curve, using the function perfcurve()

in the matlab example, the input 'scores' is from a logistic regression model scores = mdl.Fitted.Probability;

which I believe is the probability that the model gets in right in relation to the label

But how do you work out scores = mdl.Fitted.Probability; from the plsregress() function?

plsregress() does give YS, which are scores, but i'm not a mathematician, and these do not appear to be 'probabilities' in the sense that perfcurve() will take as an input

i've calculated the 'scores' another way:

yfitPLS = [ones(52,1) amidenormMaster1(:, 1:400)]*beta;
plot(group2wtTG,  yfitPLS, 'x')

which gives the graph attached

and from this I have calculated how different the model result is from the label (e.g. 1 [or 0] - 'the model result' ) and used an absolute number to get rid of negative values - but this feels wrong

what should I be doing?

EDIT: yes i'm sure how i'm calculating scores is completely wrong -

I guess another way of asking this questions is: which column of YS do I use as the 'scores', if I have fitted with 8 components (is it the 8th column?)

thanks

Charlie