B1394
Title: Bayesian semi-supervised hidden Markov models for animal movement
Authors: Vianey Leos Barajas - University of Toronto (Canada) [presenting]
Abstract: Hidden Markov models (HMMs) provide a flexible framework to model time series data where the observation process $Y[t]$ is taken to be driven by an underlying latent state process $Z[t]$, assumed to be a finite-state Markov chain. Applied to the study of animal movement, HMMs assume that the movements observed by an animal are the realizations of the animal's underlying (often unobserved) behavior. In this sense, the number of states chosen is hoped to reflect the distinct numbers of behaviors that are able to be identified from the time series alone. However, in a fully unsupervised framework, the states can be, at best, taken to be proxies of the underlying behavioral process, and, at worst, a flexible framework to capture structure in the data without connecting to biological reality. To overcome this difficulty, we can move toward a semi-supervised framework where parts of the time series are labeled in a manner to validate the behavioral processes captured. Here we discuss how different labeling designs can improve both parameter and state estimation for HMMs applied to animal movement data.
