In sci.stat.consult Phil Sherrod <[EMAIL PROTECTED]> wrote: > > On 5-May-2004, [EMAIL PROTECTED] (AJ) wrote: > >> I am trying to forecast customer life span for a set of data. >> >> Basically, we have 8 years data and thousands of rows regarding a >> subscription service. Three raw variables are as follows. >> >> a) Starting Date of subscription >> b) Cancellation Date of subscription >> c) Demograhpic Segments that a customer belongs to. We have 66 >> categorical values such as 01, 02..etc. These segments are given to >> us by an outside firm that basically appends a segment to a customer >> data based on variables such as what kind of car a customer drives, >> how much she is educated, or how much she earns etc. >> >> I am interested in predicting the number of months a customer would >> stay with the product. I was thinking I could use the following >> variables in my regression model. > > This is a good example of a data mining problem that could be handled well > by a decision tree (regression tree). Unlike classical (numeric function) > regression where your categorical variables have to be recast as multiple > binary (0/1) variables, decision trees handle categorical variables in a > natural way. I would just dump all of the data with all of the variables > into the analysis and let it pick out which variables are significant and > look for interactions. Unless there is something unusual about your data, I > believe the entire setup and analysis run could be done in a half hour. > > I recommend first developing a single-tree model which is excellent for > getting a visual picture of the model and looking for significant variables > and interactions. Then, for significantly increased accuracy, I would build > a TreeBoost model consisting a series of boosted trees. TreeBoost typically > has comparable accuracy to neural networks.
I have to chime in here and say that I find this assertion quite surprising. My admittedly limited impression of NN so far is that when put to the test of generalization/replication, they do no better than classical algorithms, save the very occasional hidden non-linearity or two. I honestly can't imagine how a tree technique isn't subject to the same limitations regarding inference as traditional models--what you suggest sounds to me like a perfect recipe for a model that is well-fitted to the sample, but has little chance of having much to do with the population from which the sample is drawn. Just because the technique doesn't use traditional concepts such as standard errors doesn't mean that they don't apply. I'd be happy to be proved wrong. Do you have any data on the success of what you suggest? Mike Babyak . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
