Message-ID: <6731304c0902260514u24096bcfi1cff1712d7014d49@mail.gmail.com>
Date: 2009-02-26T13:14:31Z
From: Max Kuhn
Subject: statistical significance of accuracy increase in classification
In-Reply-To: <BAY130-W45FE069F100B730B68A5B5C3AC0@phx.gbl>
> Do you know about any good reference that discusses kappa for classification and maybe CI for kappa???
I don't, but googling on kappa and confusion matrix etc should get you
there. Kappa works very well when the true classes are skewed. For
example, if 10% of you samples are class A and 90% class B, you can
optimize accuracy by calling everything B.
Max