You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mahout.apache.org by Lance Norskog <go...@gmail.com> on 2010/12/13 02:55:33 UTC
Precision-Recail v.s. ROC/AUC
And, also interesting for Mahout. Claims Precision-Recall is more
useful than AUC for heavily-skewed sample data.
The Relationship Between Precision-Recall and ROC Curves
http://mark.goadrich.com/articles/davisgoadrichcamera2.pdf
Jesse Davis
Mark Goadrich
--
Lance Norskog
goksron@gmail.com
Re: Precision-Recail v.s. ROC/AUC
Posted by Ted Dunning <te...@gmail.com>.
Specialized quality metrics are often necessary. Area under a small part of
the ROC curve is sometimes used.
The major advantage of AUC, however, is that you can estimate it with an
online algorithm. That is exceedingly difficult with Precision/Recall based
metrics.
Practially speaking, I have found that even with severely skewed data, AUC
is a very nice rough cut metric.
On Sun, Dec 12, 2010 at 5:55 PM, Lance Norskog <go...@gmail.com> wrote:
> And, also interesting for Mahout. Claims Precision-Recall is more
> useful than AUC for heavily-skewed sample data.
>
> The Relationship Between Precision-Recall and ROC Curves
> http://mark.goadrich.com/articles/davisgoadrichcamera2.pdf
> Jesse Davis
> Mark Goadrich
> --
> Lance Norskog
> goksron@gmail.com
>