Hi!
I need to evaluate different algorithms, and stumbled upon the AUC50 and pAUC measurements, that both quantify the results of a ROC curve.
The AUC50 (or generally, AUCx) calculates the Area under the curve up to the first 50 (x) negative examples (assuming that the results were ranked in some way, e.g. their p-value). the pAUC calculates the Area under the curve between two different settings of the FPR (or TPR).
Now my dataset has 3k positive and 7k negative examples. I am mainly interested in areas with low FPR. Thus I'd either investigate the AUCx or the pAUC from FPR 0 to some other value.
Now my question: isn't the AUC30 not the same thing as the pAUC between an FPR of 0 and 0.3? (Because at FPR = 0.003, I have 30 false negatives)? Or did I misunderstand something?
EDIT: changed "FDR" to "FPR.
