How to find Logistic Regression coefficient using SSAS (data mining)

Who is online?  0 guests and 0 members
Home  »  Forums   »  microsoft business intelligence   »  data mining   » How to find Logistic Regression coefficient using SSAS (data mining)

How to find Logistic Regression coefficient using SSAS (data mining)

Topic RSS Feed

Posts under the topic: How to find Logistic Regression coefficient using SSAS (data mining)

Posted: 7/19/2011

Jedi Youngling 2  points  Jedi Youngling
  • Joined on: 7/19/2011
  • Posts: 1

On querying a logistic regression datamining model, is it possible to find out the coefficient? Its urgent please let me know.

Thanks

Sona


Posted: 11/21/2011

Padawan 855  points  Padawan
  • Joined on: 6/7/2011
  • Posts: 46

http://msdn.microsoft.com/en-us/library/cc645906.aspx#bkmk_Query3

By using the VALUETYPE enumeration, you can return only the coefficient for the regression equation, as shown in the following query:

SELECT FLATTENED MODEL_NAME,
    (SELECT ATTRIBUTE_VALUE, VALUETYPE
     FROM NODE_DISTRIBUTION
     WHERE VALUETYPE = 11) 
AS t
FROM LR_PredictIncome.CONTENT

Posted: 1/24/2012

Jedi Youngling 30  points  Jedi Youngling
  • Joined on: 12/18/2009
  • Posts: 3

I have a Logistic Regression implementation that uses demographics and historical data to predict future action based on about a half dozen input variables. 

I'm able to retrieve the data from the Content queries, but I'm having trouble trying to tie coefficients back to anything intuitive or useful to the business in determining if the model makes sense or not.  I would like to translate coefficients into a meaningful scoring of some type and maybe even derive the complete prediction formula that leads to the probablility prediction value.  It would be great if I could hand someone a formula and say "here's how this works, plug in some values and try it".  I'm not a statastician, so I'm looking for some fairly simple reference to help me.  Any suggestions?

A particular problem with coefficients is that frequently for a given input attribute influencing a "will" or "won't" outcome, coefficients for both outcomes are negative.   How can attributes have a negative influence on both the "will" and "won't" outcomes?  I wish they simply added up to 100% or were more intuitive in some way like the Support distributions are.  The Lift Chart looks good and empirically the model is definitely working.

 


Page 1 of 1 (3 items)