ABSTRACT
As illustrated by its prevalence throughout this work, predictive modeling is a cornerstone method in database marketing. The reason is straightforward: companies can strongly benefit from deploying stored consumer information to infer future behavior or characteristics of their consumers and, by using these predictions, making marketing campaigns more effective. Strong model performance is vital, as a direct link exists between the accuracy of the predictions and the profitability of marketing campaigns. As Neslin et al. (2006) concluded from their analysis of the results of the Teradata Center for CRM and Duke University churn modeling tournament organized in 2002: methods do matter. Consequently, a vast number of studies on predictive modeling in database marketing have been devoted to introducing and benchmarking (novel) algorithms. Techniques that have been suggested in literature throughout the years include statistical techniques (for example, logistic regression (Smith, et al. 2000), generalized additive models (GAMs) (Coussement, et al. 2010), discriminant analysis (Ganesh, et al. 2000)) and classifiers originating from the data mining literature (for example, neural networks (Mozer et al. 2000), support vector machines (Coussement and Van den Poel, 2008) and decision trees (Smith, et al. 2000)).
