How to generate probabilities using data mining?

[some_math_guy]

Hi everybody,

At the telecommications provider company where I work, we use Logistic Regression (PROC LOGISTIC in SAS) models to generate a probability from 0 to 1 for each customer to buy various TelCom services (Phone, TV, High Speed Internet, Long Distance, etc) over the next 3 months, based on about 400 input feature variables (purchase history, demographics, etc). Then, whenever the sales team is running a new campaign, we will provide a list of prospective customers that are most likely to purchase the service by sorting our eligible customer list descending by probability (so that our agents call the most likely customers at the top of the list first).

This method works fine, but I would like to improve the models by using more advanced data mining techniques that I learned in graduate school such as decision trees and rule-based methods based on unsupervised machine learning (C5.0). The problem is that these models do not produce a probability score, rather they produce a decision-tree or rule set which allows you to predict some outcome for a customer by running them through the decision-tree (ie. 'will buy', 'will not buy').

I am wondering if these methods can be adapted to produce a probability of some outcome, rather than a decision-tree, so that we can 'drop in' these models in place of the old processes?

I am also wondering if people think that Logistic regression is a good method for developing predictive models for human behavior, or if they have found other methods superior.

Thanks everybody!
some_math_guy