Since the two Precision-Recall curves are crossing, I cannot decide which one is better. How should I interpret this?
1 Answer
The best model depends on what you need it for. Is your number one priority high recall or high precision? In the former case, blue seems best (*) and in the latter case the green one is better.
(*) that blue PR curve looks suspicious because you have a very long line segment there (from recall ~0.3 to 0.95). Are you sure that curve is correct?
-
$\begingroup$ What are the situations that the straight line occurs? Is it impossible? I'm not familiar with this topic of PR. $\endgroup$ Jul 17, 2015 at 2:01
-
$\begingroup$ One situation might be that you have a data point around Precision 0.45 and Recall 0.35, and then another data point around Precision 0.0, Recall 1.0, for some reason, and your plotting program is simply drawing a straight line between them. (Or there are no Recall percentages greater than 0.35.) The other is that there is a constant tradeoff between Precision and Recall over more than half of the Recall values, which seems unusual. $\endgroup$– WayneMar 4, 2018 at 16:59