MKTG 3596 Lecture Notes - Lecture 8: Nodeb, The Algorithm, Recursive Partitioning
Document Summary
A way to identify key predictors when we don"t have strong expectations about what drives customers" behavior. Most helpful when you have several possible predictors and no expectations of which is most likely to shape the behavior you want to study. Tests lots of different possible predictors of an outcome. Choose combination that best fits the data. Apply your best combination to new data to see how well it works. As an end in itself: stratification/classification/segmentation. Assign customers into different categories to decide who your target should be for the next mailing, etc. As input to another model: data reduction/variable screening: identify a useful subset of predictors from a large set - for use in a model such as regression or logit. As input to another model: interaction identification: identify relationships that pertain only to specific subgroups for use in a model such as regression or logit.
