ROC_AUC curve for performance of a classification

Question

[twinklemishra0892]

I am doing LEfSe analysis between three sample types(control, asymptomatic and symptomatic) in genus level taxa. And from LEfse results, taken all those bacteria and trying to do ROC AUC to determine classification performance. when compared symptomatic and asymptomatic groups, getting fair AUC percentage (76.6%) but when trying to compared with control group getting AUC 100%.I want clarify on this point, is it possible to get AUC 100%

Answer 1

[Kevin Blighe]

I want clarify on this point, is it possible to get AUC 100%

It is possible but not credible. This can occur when the model is 'overfit' and there is a low sample number.

Please clarify the sample numbers.

Kevin

[twinklemishra0892]

Total 58 samples (12 control, 25 Asymptomatic and 21 Symptomatic)

[Kevin Blighe]

The imbalanced groups, perhaps. One would have to review every step in fine detail to understand what was happening, though.