Work

MDM: An R package for causal multivariate time series tasks

Year: 2025

Type: article

Abstract: The Multiregression Dynamic Model (MDM) is a framework that combines graph theory with dynamic linear models, allowing a non-Gaussian multivariate structure to emerge in the context of causal time series. Since an optimal DAG structure is an NP-hard task, this package overcomes the all-combinations search (Integer Programming Algorithm) using heuristic algorithms (like Hill Climbing). Written using R S4 Object programming, it performs learning functions (estimating network structure and its dynamic arcs), as well as includes DAG (causal) visualization, time-varying coefficients visualization, and graphical performance checks. The MDM R package is distributed under the GPL license and is accessible from GitHub. • Flexible Time Series model for non-stationary and time-varying parameters. • Bayesian dynamic model via Kalman Filter/Smooter closed form. • Multivariate Time Series estimation via probabilistic graphical model. • Dynamic Bayesian Networks pattern discovery using Heuristic-algorithm.

Source: Software Impacts

DOI: 10.1016/j.simpa.2025.100796

Authors: Lilia Costa, Arthur Azevedo, Michel Miler Rocha dos Santos, Diego Carvalho Nascimento

Institutions: Universidade Federal da Bahia, NEOMA Business School

Language: English

Topic: Bayesian Modeling and Causal Inference

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $600

Funder: Conselho Nacional de Desenvolvimento Científico e Tecnológico