Many times in the literature an area under a ROC curve (AROC or AUC) is reported without an indication of what it means.
General rules:
(1) A test with a greater AROC is "better" than a test with a lower AROC.
(2) A test with an AROC of 1.0 is a "perfect" test.
(3) A test with an AROC <= 0.5 is "useless", maybe even harmful.
With this in mind the following quality correlation is proposed.
|
AUC or AROC
|
Performance
|
|
< 0.50
|
extremely poor
|
|
0.50 - 0.549
|
very, very poor
|
|
0.55 - 0.599
|
very poor
|
|
0.60 - 0.649
|
poor
|
|
0.65 - 0.699
|
subpar- to-poor
|
|
0.70 - 0.749
|
subpar
|
|
0.75 - 0.799
|
fair
|
|
0.80 - 0.849
|
fair-to-good
|
|
0.85 - 0.899
|
good
|
|
0.90 - 0.949
|
very good
|
|
0.95 - 0.999
|
excellent
|
|
1.00
|
perfect
|
Limitation:
• The progression in the AROC is linear, while the quality may be nonlinear.
