Research Project Title:
Variable Selection in Treatment Effect Prediction
abstract:Even though machine learning models trained to predict the treatment effect are very well developed, problems still happen in the context of decision making based on the past observed data. In many cases, the observed data is drawn from the treatment groups, while we try to predict the treatment effect on the control groups, which have different distribution. Research areas such as the variable selection to preserve unconfoundedness or the identification of the common support of the observed and unobserved data then arise. We will start with the use of decision-base algorithms to aid the selection of variables, with the potential of moving on to other tools and models. Once the models are built and proven, we will apply them to the data in medical context such as Electronic Health Records.
Unlike classes with problem sets and small projects, SuperUROP gives me an opportunity to do a full scale project with both technical and communication guidance. It will also help me get the feeling of studying for a doctorate. I believe that this machine learning in causal inference project will be a good blend of knowledge I have got from classes in Machine Learning, Inference, and Econometrics.