This webinar will introduce learners to the basic principles of machine learning (ML) for prediction in health economics and outcomes research (HEOR).
It will begin by framing the prediction problem and drawing both connections and contrasts with the related problems of explanation and causal inference. It will then confront the “Iron Law of Prediction” – which characterizes the goals of prediction in terms of “minimizing loss” – or the consequences of making poor predictions. The webinar will then explore the principles of “cross validation” and the “tuning” of algorithms to minimize loss. Using a simulated dataset, the presenter will apply several ML algorithms to illustrate the principles of “complexity penalization,” “bagging,” “random feature selection,” and “boosting.”
The webinar will conclude by considering an empirical application of boosting to construct propensity scores – a method of prediction that is used to improve causal inference for other parameters of interest.
This webinar will use a combination of mild mathematical formalization (yes, there will be equations) and visualization. While advanced training in statistics, epidemiology, or econometrics is not required, a basic comfort with statistical concepts will help learners get the most out of the webinar. The R code used to generate and analyze the simulated dataset will be available for download.