A0408
Title: A Bayesian nonparametric spiked process prior for dynamic model selection
Authors: Michele Guindani - University of California, Irvine (United States) [presenting]
Abstract: In many applications, investigators consider processes that vary in space and time, with the goal of identifying temporally persistent and spatially localized departures of those processes from a baseline or ``normal" behavior. We propose a Bayesian nonparametric model selection approach for the analysis of spatio-temporal data, which takes into account the non-exchangeable nature of measurements collected over time and space. More specifically, a zero-inflated conditionally identically distributed (CID) species sampling prior is used to model temporal dependence in the selection, by borrowing information across time and assigning data to clusters associated to either a null or an alternate process. Spatial dependences are accounted for by means of a Markov random field (MRF) prior, which allows to inform the selection based on inferences conducted at nearby locations. We investigate the performances of our model by means of a simulation study and an application to a disease surveillance problem, for detecting outbreaks of pneumonia and influenza (P\&I) mortality in the continental United States. We show how the proposed modeling framework compares favorably with respect to commonly adopted threshold methods for detecting outbreaks over time and also to recent proposals modeling more complex Markov switching dependences.
