Title

Utilizing Uplift Modeling to Develop a Credit Line Increase Strategy

Presenter Information/ Coauthors Information

Sebastian SowadaFollow

Presentation Type

Poster

Student

Yes

Track

Finance/Insurance Application

Abstract

Traditionally, response models have been used to target individuals who are most likely to respond (yes/no, good/bad) to a direct action (marketing campaign, credit line increase). However, response models are incapable of predicting who is most affected by the direct action. In this paper we will analyze the practice of uplift modeling, the predictive modeling technique used by statisticians to measure the incremental effect of a treatment. We will examine the theoretical background of two types of uplift modeling: difference in scores, and uplift trees. This paper will then describe different validation approaches used specifically to uplift modeling, such as the Qini statistic and Net information value. This paper will conclude in a case study using data on credit card customers. We will use uplift modeling to predict customers who are most likely to have a positive incremental impact from a credit line increase. Finally, using both types of uplift modeling, we will compare the uplift brand of models to a traditional response model and detail why uplift modeling is more appropriate in this situation.

Keywords: Uplift, Predictive Modeling, Credit Line Increase

Start Date

5-2-2019 12:00 PM

End Date

5-2-2019 1:00 PM

This document is currently not available here.

Share

COinS
 
Feb 5th, 12:00 PM Feb 5th, 1:00 PM

Utilizing Uplift Modeling to Develop a Credit Line Increase Strategy

Volstorff A

Traditionally, response models have been used to target individuals who are most likely to respond (yes/no, good/bad) to a direct action (marketing campaign, credit line increase). However, response models are incapable of predicting who is most affected by the direct action. In this paper we will analyze the practice of uplift modeling, the predictive modeling technique used by statisticians to measure the incremental effect of a treatment. We will examine the theoretical background of two types of uplift modeling: difference in scores, and uplift trees. This paper will then describe different validation approaches used specifically to uplift modeling, such as the Qini statistic and Net information value. This paper will conclude in a case study using data on credit card customers. We will use uplift modeling to predict customers who are most likely to have a positive incremental impact from a credit line increase. Finally, using both types of uplift modeling, we will compare the uplift brand of models to a traditional response model and detail why uplift modeling is more appropriate in this situation.

Keywords: Uplift, Predictive Modeling, Credit Line Increase