B1572
Title: Quantile functional regression for high dimensional data streams
Authors: Jeffrey Morris - University of Pennsylvania (United States) [presenting]
Ye Emma Zohner - Rice University (United States)
Abstract: A new methodology is presented for modeling data streams in a Bayesian functional regression framework. The general strategy is to partition the data stream into serial epochs, compute the distribution of observations within each epoch that we represent in a custom quantile function space, and model as a functional object. We assess how the distributions vary over time and over subject-specific covariates. The motivating data for this methodology is intraocular pressure data collected from sensors placed on the eyes of non-human primates. This study characterizes intraocular pressure which is linked to glaucoma, a group of eye diseases that affects millions of people worldwide. Although our methodology is motivated by intraocular data streams, the methods we develop can broadly be used for functional regression on data streams to assess how subject-level and temporal factors affect the dynamic shifting distribution of the data represented in the stream.
