You are here
Theoretical and Applied Data Science Lunch-n-learn - Chancellor Johnstone
Presenter: Chancellor Johnstone
Title: Distribution Free Prediction Intervals
Abstract: We deliver a novel distribution free prediction interval methodology based on the ranking method introduced in Harville 1977. Typical prediction intervals are based on estimates of the mean and variance of an observation and include assumptions on the normality, zero expectation, and constant variance of error terms. We instead focus our assumptions on the consistency of the estimated model parameters. With this, we build an increasing collection of modified residuals coming from periodic, out-of-sample predictions that approximates the true error distribution for the outcome of interest. We apply this methodology to case studies across multiple sports, and image ranking analysis.
Bio: Chance Johnstone is a PhD student in the Department of Statistics at Iowa State University, working under Dr. Dan Nettleton. His current research focuses on nonparametric and distribution free methods. He will be part of the faculty at the Air Force Institute of Technology in Dayton, Ohio beginning in August. In his free time he likes to rock climb, watch Soccer, go on bike rides with his wife, Rachel, and go on runs with his dog, Pepper.
After the presentation, there will be a short time for discussion and questions afterwards.