Loading...

Media is loading
 

Paper Number

2313

Paper Type

Complete

Description

The design of targeting policies is fundamental to address a variety of practical problems across a broad spectrum of domains from e-commerce to politics and medicine. Recently, researchers and practitioners have begun to predict individual treatment effects to optimize targeting policies. Although different research streams, that is, uplift modeling and heterogeneous treatment effect propose numerous methods to predict individual treatment effects, current approaches suffer from various practical challenges, such as weak model performance and a lack of reliability. In this study, we propose a new, tree- based, algorithm that combines recent advances from both research streams and demonstrate how its use can improve predicting the individual treatment effect. We benchmark our method empirically against state-of-the-art strategies and show that the proposed algorithm achieves excellent results. We demonstrate that our approach performs particularly well when targeting few customers, which is of paramount interest when designing targeting policies in a marketing context.

Comments

13-DataAnalytics

Share

COinS
 
Dec 12th, 12:00 AM

The Best of Two Worlds – Using Recent Advances from Uplift Modeling and Heterogeneous Treatment Effects to Optimize Targeting Policies

The design of targeting policies is fundamental to address a variety of practical problems across a broad spectrum of domains from e-commerce to politics and medicine. Recently, researchers and practitioners have begun to predict individual treatment effects to optimize targeting policies. Although different research streams, that is, uplift modeling and heterogeneous treatment effect propose numerous methods to predict individual treatment effects, current approaches suffer from various practical challenges, such as weak model performance and a lack of reliability. In this study, we propose a new, tree- based, algorithm that combines recent advances from both research streams and demonstrate how its use can improve predicting the individual treatment effect. We benchmark our method empirically against state-of-the-art strategies and show that the proposed algorithm achieves excellent results. We demonstrate that our approach performs particularly well when targeting few customers, which is of paramount interest when designing targeting policies in a marketing context.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.