question about auroc for plsda

First of all, thanks a lot to you and your team for such a wonderful R package.

I have a question, though, regarding the ROC curve, and the value of outcome represented in the plot (I attach an example, where S and M are my two groups)

  • First, what is the outcome shown in the plot? Accuracy? how is it decided?

  • second, if this value is computed using cross-validation, what are the default values of the folds and the nrepeats?

I have read some documentation were it states:

An AUC plot can also be obtained using the function auroc, where the AUC is calculated from training cross-validation sets and averaged (see the perf function outputs for each component, perf.plsda.srbct$auc and…

and here, were it states:

As PLS-DA acts as a classifier, we can plot a ROC Curve to complement the sPLS-DA classification performance results detailed in 4.7.5. The AUC is calculated from training cross-validation sets and averaged.

Hi @janis_pi

We can’t really see your output but basically the ROC is calculated based on the predicted components to predict the class of each sample, when doing cross-validation (the prediction process is described in details in the supplemental of I think this is what you mean by ‘outcome’?
As you know, a ROC outputs the spec / sens of the prediction for every possible cutoff value of a measure (here predicted component). However in PLS-DA we do say that it does not make much sense, since PLS-DA has its internal cutoff for class prediction using the prediction distances. So it is almost as if we are going backward, or kind of ignore the prediction of PLS-DA anyway. :neutral_face:

The folds / repeats are the same as from the perf() function you have run to obtained the AUC, auroc() applies to a perf() object.