Marek Brabec

Dept. of Statistical Modeling & Artificial Intelligence

Institute of Computer Science of the Czech Academy of Sciences

Location: Institute of Computer Science CAS, Pod Vodárenskou věží 2, 180 00 Prague, Czechia

   Room 318

Date:   Wednesday 25 November 2026

Time: 14:00 CET

Abstract:

In this talk, we present an approach to the dynamic modeling of multivariate time series based on a state-space model, and thus on a decomposition into a regression component, a latent dynamic component, and an observation error based on prior knowledge. We demonstrate a Bayesian implementation in the case of relatively short series observed independently across a larger number of subjects—both for individual modeling and for a hierarchical model.

Modeling both of these types finds application in the analysis of EMA (Ecological Momentary Assessment) data on human population mobility. While the hierarchical approach is advantageous for analyzing and understanding the structure of both dynamic and exogenous dependencies, individual models are applied in short-term predictions useful for the timing of active interventions that support physical activity.

We illustrate the use of the state-space approach using data from the extensive research program OP JAC DigiWell (Z.02.01.01/00/22_008/0004583).

Keywords: Multivariate dynamic model, Bayesian modeling.

References: 

1. Brabec,M.-Jílek,K. (2012): Dynamic Assessment of Radon Source in Buildings, Based on Tracer Gas Experiment Statistical Modeling. Chapter 8 in: Handbook of Radon: Properties, Applications and Health (Li,Z. and Feng,C., eds.), Nova Publishers, ISBN: 978-1-62100-177-5

2. Froňka,A.-Jílek,K.-Moučka,L.-Brabec,M. (2011): Significance of independent radon entry rate and air exchange rate assessment for the purpose of radon mitigation effectiveness proper evaluation: case studies. Radiation Protection Dosimetry, 145, 2-3, 133-137