On 25 Feb 2004 at 5:06, Robert Chapman wrote:
Someone calls what you described for data mining. More specifically you could use logistic regression or a regression tree, with helper methods like cross-validation. You said price is not a concern, so I assume zero price is OK? Then google for CRAN (Comprehensive R Archival Network), and download R for windows. You will need some information to get you started, there is some free downloadable books on the CRAN site, and some published books, mentioned in the R FAQ. If you are not a programmer, it will take some time getting started, but not as much as with SAS! Kjetil Halvorsen > Hi, > > I am working on a company initiative to replace an external > statistical modelling contract with the internal resources to handle > the same. I would be very grateful for any recommendations of the > appropriate software we need to acquire for this purpose. I'll > describe the sort of modelling we are doing: > > We are trying to predict the behaviour of stores (customers), this > behaviour being binary - either they do what I'll call 'x' or they do > not. Sampling has been done to determine which stores do and which > don't and we have database information on all these sample stores (as > well as all our other stores). What we want to do is predict x using > all the information in our databases so as to be able to target a > reduction of x. If we randomly select stores to speak with them we > have a 10% chance of them being a store that does x. We'd like to > increase this percentage using a model that produces a decision tree > based on all the factors such that stores get grouped according to > type , e.g. if factor "A" >10,000 Then.......If factor "B" >35 days > then a store is in category 17 which is likely to exhibit x. With > this sort of modelling you would need to be able to specify weightings > since there is a tradeoff between identifying every store with > behaviour x and the accuracy of prediction. > > I assume we simply need some software that handle multiple regression > very capably. But perhaps there is a more specific description of > what we are trying to do. The external contractor we are using at the > moment is using SAS but we need something Windows-based (not Unix). > Price is not really a factor and we will only use it for the data > analysis described. > > So I have two questions really - what do you call the statistical > process I have described and what software would be recommendable to > use for it? > > Appreciate any advice, > > Robert > . > . > ================================================================= > Instructions for joining and leaving this list, remarks about the > problem of INAPPROPRIATE MESSAGES, and archives are available at: . > http://jse.stat.ncsu.edu/ . > ================================================================= . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
