New Uplift Modelling Theory Paper Available
There is a new paper available for download discussing the theory of uplift modelling. It is available here.
This paper, for the first time, details the algorithm that has formed the basis for the uplift models built by the product now known as Portrait Uplift, from Pitney Bowes, and which was originally part of the Decisionhouse software from Quadstone Limited. These uplift models are based on one or more decision trees built using modified split criteria and alternative pruning methods. Because the modified split criterion is based on a significance test, we call the trees Significance-Based Uplift Trees.
The paper also details many of the insights that the Pitney Bowes Business Intelligence team and I have gained over some twelve years of building commercial uplift models, and was written jointly with my long-term collaborator, Patrick Surry.
The paper is called Real-World Uplift Modelling with Significance-Based Uplift Trees and the abstract is below. We hope to publish it in a (peer-reviewed) publication, perhaps in modified form.
This paper seeks to document the current state of the art in ‘uplift modelling’—the practice of modelling the change in behaviour that results directly from a specified treatment such as a marketing intervention. We include details of the Significance-Based Uplift Trees that have formed the core of the only packaged uplift modelling software currently available. The paper includes a summary of some of the results that have been delivered using uplift modelling in practice, with examples drawn from demand-stimulation and customer-retention applications. It also surveys and discusses approaches to each of the major stages involved in uplift modelling—variable selection, model construction, quality measures and post-campaign evaluation—all of which require different approaches from traditional response modelling.