Boosting algorithms for uplift modeling

TL;DR

This paper adapts boosting algorithms for uplift modeling, enabling better prediction of causal effects of actions like marketing or medical treatments, with three proposed algorithms demonstrating significant performance improvements.

Contribution

It introduces three novel uplift boosting algorithms, each satisfying different desirable properties, advancing the methodology for causal effect prediction in uplift modeling.

Findings

Proposed algorithms outperform base models significantly

All three algorithms satisfy two of the three desirable properties

Experimental results confirm the effectiveness of the methods

Abstract

Uplift modeling is an area of machine learning which aims at predicting the causal effect of some action on a given individual. The action may be a medical procedure, marketing campaign, or any other circumstance controlled by the experimenter. Building an uplift model requires two training sets: the treatment group, where individuals have been subject to the action, and the control group, where no action has been performed. An uplift model allows then to assess the gain resulting from taking the action on a given individual, such as the increase in probability of patient recovery or of a product being purchased. This paper describes an adaptation of the well-known boosting techniques to the uplift modeling case. We formulate three desirable properties which an uplift boosting algorithm should have. Since all three properties cannot be satisfied simultaneously, we propose three uplift boosting algorithms, each satisfying two of them. Experiments demonstrate the usefulness of the proposed methods, which often dramatically improve performance of the base models and are thus new and powerful tools for uplift modeling.