YSPH Biostatistics Seminar: "Similarity-based Statistical Modeling"

Attendance via Zoom.

Speaker- Joel Dubin, Ph.D.

Title- Similarity-based statistical modeling

Abstract

Similarity is a commonly used term in statistics, including for purposes of matching study participants or for either unsupervised or supervised learning contexts. In this presentation, we will look at some uses of similarity for the aim of improving predictive model accuracy when utilizing statistical modeling in health studies. Thought needs to go into how to define similarity, including consideration of an unsupervised or supervised approach; how to incorporate categorical predictors; how to utilize weighting; how to tune similarity; and how to quantify uncertainty when implementing a similarity step. Some applications with intensive care unit data will be discussed.