Bounded Individualized Treatment Effect: A Novel Approach for Uplift Modeling

The advent of artificial intelligence has greatly facilitated the prediction of individualized treatment effects, expanding the field of causal inference beyond the traditional realm of econometric studies. This shift has enabled prescription of personalized treatments via predictive analytics, allowing businesses to deliver personalized and relevant interactions with customers across various touchpoints. In this paper, we propose a novel approach for uplift model estimation called Bounded Individualized Treatment Effect (BITE), which is designed for business analytics tasks. The goal is to prioritize customers who will be persuaded by the treatment, by filtering individuals based on their baseline probabilities. Our experiments show the virtues of the BITE approach on seven benchmark datasets and 21 uplift classifiers, achieving best average performance compared to traditional uplift modeling and standard predictive classification.