Title: Profit driven decision trees for churn prediction
Authors: Sebastiaan Hoeppner - KU Leuven (Belgium) [presenting]
Eugen Stripling - KU Leuven (Belgium)
Bart Baesens - KU Leuven (Belgium)
Seppe vanden Broucke - KU Leuven (Belgium)
Tim Verdonck - UAntwerp, KU Leuven (Belgium)
Abstract: The interest for data mining techniques has increased tremendously during the past decades, and numerous classification techniques have been applied in a wide range of business applications. In the telecommunication sector, companies heavily rely on predictive churn models to detect churners in a vast customer base and to remain competitive in a saturated market. In a recent paper, the expected maximum profit (EMP) has been proposed which explicitly takes the cost of offer and the customer lifetime value of retained customers into account. It thus permits the selection of the most profitable classifier which better aligns with business requirements of end-users and stake holders. However, modelers are currently limited to applying this metric in the evaluation step. Therefore, we present a classifier named ProfTree, that maximizes the EMP metric in the training step using a genetic algorithm. The technique is based on a classification tree for modeling a binary response variable.