Ensemble methods in bank direct marketing

Abstract

Increasing costs of direct marketing campaigns and declining response rates have motivated direct marketers to turn to more sophisticated techniques to model response behavior. Moreover, the data used for response modeling is imbalanced data. That is, non-respondents greatly outnumber respondents in direct marketing. This paper intends to compare bagging with boosting algorithms to check how well these methods perform when class imbalance problem occurs in bank directing marketing data.