Bayesian formulation and implementation for a non-stationary semi-Markov model with covariates

by Sebastien Coube (Inrae, Unité MIAT)

Tuesday Nov 4, 2025, 11:15 AM → 12:15 PM Europe/Paris

Salle K. Johnson (1R3, 1er étage)

Modeling the impact of co-variates upon the transition from one hidden state to another may improve both estimation and prediction in hidden hidden Markov and semi-Markov models. However, the problem of the interpretability of the parametrization and feasability of the computation becomes crippling as the number of interest covariates increases. I propose, in a Bayesian perspective, an architecture I hope to be as complex as needed to accommodate complex configurations of the data, but as simple as possible for the sake of usability. A few simulated examples are given to illustrate the flexibility of the model. A Bayesian strategy to fit the model is discussed.

(Work in progress)