Date: Sat, 17 Dec 2005 10:03:42 -0500
Reply-To: Mike Donatello <MDonatello@cox.net>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Mike Donatello <MDonatello@cox.net>
Subject: Judging improvement in fit for CART-based models
Content-Type: text/plain; charset="windows-1250"
Can anyone point me to a test statistic for judging the improvement in fit
between two CART-derived models? For instance, if Model 1 gives me 65%
correct classification and Model 2 gives me 72% correct classification, how
do I judge whether the increase in classification accuracy is significant?
And, how is that significance affected by the number of predictors included?
If there is no established fit statistic, would testing for the significance
in proportion of correct classifications be appropriate?
I have several texts which discuss CART details, including the original
Breiman et al. work, but I canít find an answer in any of them.
No virus found in this outgoing message.
Checked by AVG Free Edition.
Version: 7.1.371 / Virus Database: 267.14.1/206 - Release Date: 12/16/2005