Date:         Tue, 4 Oct 2005 07:52:26 -0400
Reply-To:     Rajat Mathur <RMathur@INDUCTIS.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Rajat Mathur <RMathur@INDUCTIS.COM>
Subject:      Doubt on Prediction using Clustering
Comments: cc: Amanjeet Saluja <asaluja@Inductis.com>,
          Neha Swetambari <NSwetambari@inductis.Mi8.com>,
          Meesum Kazmi <MKazmi@Inductis.com>
Content-Type: text/plain; charset=us-ascii

Hi,

I am try predicted a variable with help of some independent variables. I
have made homogenous cluster by using Classification and Regression Tree
(giving target variable as my dependent variable).

Here if I choose Median/MAD ratio maximum I am getting better result
(may be my clusters are more homogenous) as compare to maximum
Average/STD.

I need to know is it fine to go with Median/MAD max or I am missing
something here. I mean how is Median/MAD is better then Average/STD.

Also in my Model after prediction (using proc reg) I am getting high
RMSE and higher MAD. But if I find the absolute difference between my
actual values and predicted values I find them better when compare to
other models where I am getting lower RMSE and lower MAD but in these
models the result of absolute difference between my actual values and
predicted values is not good.

Can anyone suggest what can be the reason for this?


Regards,
Rajat