1

Can I draw a single ROC curve for a test method applied onto a group of people of different levels of sickness severity (i.e., light, mild, severe)? Or I need to draw three different ROC curves, one for each severity level? If the answer is yes, what is this kind of ROC experiment called? Thanks

Ali Sultan
  • 563
  • 1
  • 5
  • 14
  • 1
    Might be related to this question http://stats.stackexchange.com/questions/2151/how-to-plot-roc-curves-in-multiclass-classification – Vincent Guillemot Feb 23 '16 at 17:18
  • Thanks Vincent, but that question is totally different. – Ali Sultan Feb 24 '16 at 01:36
  • I think you could give it a try: you have three classes, and this technique will allow you to draw ROC curves and compute AUCs. Sure it will not take into account the fact that your factor of interest is ordinal, but it's a start! – Vincent Guillemot Feb 24 '16 at 11:08
  • Thanks again Vincent, but I am trying to classify patients as either sick or healthy (i.e., two classes) but the patients actually of different sickness severity. Is it conceptually correct to assume that all patients are of same sickness level and draw a single ROC curve for the test method? – Ali Sultan Feb 25 '16 at 04:19
  • You need to add these important (and not obvious) details in your question. – Vincent Guillemot Feb 25 '16 at 14:52

0 Answers0