Machine learning has vast untapped potential for helping medical professionals by predicting a patient’s response to a treatment. However, many machine learning techniques are not able to effectively quantify uncertainty. This leads to unjustified recommendations that the medical industry is hesitant to adopt in practice. The goal of this project is to design an algorithmic framework for estimating individual treatment effects from observational data that incorporates sensitivity analysis from the start. We use an extrapolation-based approach to quantify regions of the population space for which our treatment effect estimates are applicable. This allows us to make treatment recommendations that are valid despite uncertainty in the causal effect.

“I am interested in this SuperUROP project because it will provide me with the opportunity to work in multiple machine learning paradigms and to put the theory I have been studying in my classes to practical use. I hope to learn the subtleties of causal inference and time series modeling of medical data while simultaneously improving my own implementation and research abilities.”