A critical review of common pitfalls and guidelines to effectively infer parameters of agent-based models using Approximate Bayesian Computation

The agent-based modelling paradigm often results in complex, highly detailed models, containing unknown or uncertain parameters. Approximate Bayesian Computation (ABC) offers a simulation-based approach for inferring these parameters from observational data. But similar to the flexibility ingrained...

Full description

Saved in:
Bibliographic Details
Published in:Environmental modelling & software : with environment data news Vol. 172; p. 105905
Main Authors: De Visscher, Lander, De Baets, Bernard, Baetens, Jan M.
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.01.2024
Subjects:
Agent-based models
Approximate Bayesian Computation
Bayesian theory
Calibration
computer software
Individual-based models
Inference
observational studies
Simulation
Approximate Bayesian Computation
Individual-based models
Inference
Agent-based models
Calibration
Simulation
ISSN:1364-8152, 1873-6726
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The agent-based modelling paradigm often results in complex, highly detailed models, containing unknown or uncertain parameters. Approximate Bayesian Computation (ABC) offers a simulation-based approach for inferring these parameters from observational data. But similar to the flexibility ingrained in agent-based models, the flexible nature of ABC involves several design choices. Here we systematically review how ABC is currently applied in combination with agent-based models, with about half of the reviewed applications being set in an ecological context. We provide a critical discussion of common practices, accompanied by illustrative examples with a benchmark model from the Agents.jl Julia package. This sets out guidelines to aid modellers that are unfamiliar with the subject in their research endeavors. •We review the application of ABC for estimating parameters of agent-based models.•We find that necessary validation methods are applied too infrequently.•Our remarks are illustrated by simulations with a benchmark model in Agents.jl.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:1364-8152
1873-6726
DOI:10.1016/j.envsoft.2023.105905