Re: Calculating Expected Value With Multiple Variables Regression
- From: "Phil Sherrod" <phil.sherrod@xxxxxxxxxxxxxxxxxxx>
- Date: Tue, 2 Aug 2005 14:11:10 GMT
On 2-Aug-2005, wrote:
> I'm working on a problem in the sales area where I like to predict the
> "worth" of a deal by looking at multiple known attributes. I like to
> do this by running regressions against historical data (e.g. "A
> prospect with 200 employees and a blue logo on the average buys $20K
> worth of product over the past 10 years").
>
> What makes this a little tricky is that the set of attributes are not
> always the same. So one deal may have 10 attributes, the other deal
> has 6, and both only have 3 attributes in common. Also I'm not only
> trying to figure out the probability of a deal happening, but also the
> deal size (e.g. $10K). I assume there is enough correlation between
> the attribute and the outcome (a sale) that the resulting analysis has
> predictive power.
You are correct that your mixed sets of attributes makes this problem
difficult. There are ways of handling missing values when performing
multiple regression, but I think you may get better results by trying a
different approach that is less depending on having all variables.
I would like to run your data through my DTREG program and try to build a
model based on a decision tree. DTREG has the ability to use surrogate
splitters to compensate for missing attributes: basically, it looks for
other variables that can be used as surrogates when there is a missing value
on one or more predictors. For example, if the value of Number Of Employees
is missing, it might determine that Annual Staff Salary is an acceptable
substitute. This is done dynamically as the model is built and different
surrogate predictors can be selected for each group as the trees are built.
If you can e-mail your data to me at phil.sherrod 'at' sandh.com, I will be
happy to see if I can build a model for it.
--
Phil Sherrod
(phil.sherrod 'at' sandh.com)
http://www.dtreg.com (decision tree modeling)
http://www.nlreg.com (nonlinear regression)
.
- Prev by Date: Re: Calculating Expected Value With Multiple Variables Regression
- Next by Date: Re: Q. about censoring in survival analysis problem
- Previous by thread: Re: Calculating Expected Value With Multiple Variables Regression
- Next by thread: Q. about censoring in survival analysis problem
- Index(es):
Relevant Pages
|
Loading