Generating scoring models with auxiliary target variables and data bridging
The authors present an innovative way to deal with a frequent practical problem in CRM and direct marketing projects: the lack of cases with known target behavior. Usually, this makes learning scoring models that can be utilized for selecting the target group of a new campaign infeasible. The approach presented resorts to auxiliary target variables whose nature is derived through analogous induction and whose similarity can be calculated by using odds ratio and Euclidian distance. This paper illustrates how the methodology works on a practical example taken from the automotive industry.
- This could also be of interest