On Friday, October 13th, 2023, the seminar series hosted a talk from Mackenzie Neal, a PhD researcher from McMaster University in Canada, discussing Hidden Markov Models for multivariate data. Details for the talk are below.
Title
Hidden Markov Models for Multivariate Longitudinal Data
Abstract
While advances continue to be made in model-based clustering, challenges persist in modeling various data types such as longitudinal data. Multivariate longitudinal data presents difficulties to clustering algorithms due to the unique correlation structure, a consequence of taking observations on several subjects over multiple time points. Additionally, longitudinal data is often plagued by missing data and dropouts, presenting issues for estimation algorithms. For certain longitudinal studies, it would be useful not only to cluster the subjects but to model the transitions between states. The change in state can be modeled by hidden Markov models (HMMs). A modified expectation-maximization (EM) algorithm is used to estimate model parameters in the presence of latent variables. This research builds an EM algorithm in the context of HMMs that addresses missing not at random (MNAR) data with a longitudinal covariance update to compensate for the unique longitudinal correlation structure.