Designing Customer Signatures
Before going into the details of the data manipulations, let’s cover some key ideas concerning the design of customer signatures. These ideas ensure that they work well for analytic purposes, and that they can be generated to be as-of arbitrary points in time.
Profiling versus Prediction
Chapter 11 introduces the distinction between a profiling model set and a prediction model set. In a profiling model set, the inputs and the targets come from the same time frame. In a prediction model set, the inputs are known before the target. The same ideas hold for customer signatures.
This chapter focuses on prediction model sets because they are more powerful. In a profiling model set, the target variables can simply be created in the same way as the input variables. In a prediction model set, the cutoff date is for the input variables, and the target comes from a time frame after the cutoff date.
Column Roles
Columns in a customer signature have various...