Categorical surrogation of agent-based models - A comparative study of machine learning classifiers

Aug 4, 2026

Upload by Oliver

Agent-based modelling supports natural exploration of social system behaviours but often becomes computationally prohibitive due to the large parameter space. Machine learning-based surrogates or metamodels can act as efficient proxies, yet clear guidance on which algorithms suit categorical metamodeling remains limited. This work delivers a systematic comparative analysis of machine learning classifiers for agent-based model surrogation, assessing correctness, efficiency, and robustness across multiple datasets and sample sizes using an artificial market case study.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages40

Reading Time1 h 41 min

AccessFree to download

Introduction

Agent-based modelling and its computational challenge

Machine-learning surrogates and the gap in categorical metamodeling

Received: 16 September 2022 Revised: 22 March 2023 Accepted: 8 May 2023 DOI: 10.1111/exsy.13342 O R IG INA L ARTI CLE Categorical surrogation of agent-based models: A comparative study of machine learning classifiers Bàrbara Llacay 1 | Gilbert Peffer 2 1Department of Business, Faculty of Economics and Business, University of Barcelona, Barcelona, Spain 2Centre Internacional de Mètodes Numèricsen Enginyeria (CIMNE), Barcelona, Spain Correspondence Bàrbara Llacay, Department of Business, Faculty of Economics and Business, University of Barcelona, Av. Diagonal 690 08034 Barcelona, Spain. Email: [bllacay@ub.edu](bllacay@ub.edu) Funding information Universitat de Barcelona, Grant/Award Number: AS017634 Abstract Agent-based modelling has gained recognition in the last years because it provides a natural way to explore the behaviour of social systems. However, agent-based models usually have a considerable number of parameters that make it computationally prohibitive to explore the complete space of parameter combinations. A promising approach to overcome the computational constraints of agent-based models is the use of machine learning-based surrogates or metamodels, which can be used as efficient proxies of the original agent-based model. As the use of metamodels of agent-based simulations is still an incipient area of research, there are no guidelines on which algorithms are the most suitable candidates. In order to contribute to filling this gap, we conduct here a systematic comparative analysis to evaluate different machine learning-based approaches to agent-based model surrogation. A key innovation of our work is the focus on classification methods for categorical metamodeling, which is highly relevant because agent-based simulations are very often validated ina qualitative way. To analyse the performance of the classifiers we use three types of indicators—measures of correctness, efficiency, and robustness—and compare their results for different datasets and sample sizes using an agent-based artificial market as a case study. KEYWOR DS agent-based model, classifier, machine learning, metamodel, surrogation 1 | INTRODUCTION Agent-based modelling is an emerging subfield of distributed artificial intelligence that focuses on the modelling and simulation of complex systems with multiple agents. In the last years, agent-based models have gained widespread recognition because they provide a natural paradigm to build and explore a wide range of so-called artificial social systems, that is, models that allow researchers to study the behaviour of real social systems. Agent-based models (ABMs) represent real systems as networks of autonomous and heterogeneous agents that interact with each other and with their environment. The basic unit that underlies an ABM is the agent. Although there is no universally accepted definition of what an agent is, one of the most widespread definitions is the one proposed by Wooldridge: “An agent is a computer system that is situated in some environment, and that is capable of autonomous action in this environment in order to meet its design objectives”(Wooldridge, 1999, p. 29) . ‘Situated’means that the agent is immersed in an environment, receives information from that environment and from other agents, and acts in ways that This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2023 The Authors. Expert Systems published by John Wiley & Sons Ltd. Expert Systems. 2025;42:e13342 . [https://doi.org/10.1111/exsy.13342](https://doi.org/10.1111/exsy.13342) [wileyonlinelibrary.com/journal/exsy](wileyonlinelibrary.com/journal/exsy) 1 of 40 2 of 40 LLACAY and PEFFER can alter this environment and the behaviour of the interacting age

FAQs

Why are agent-based models computationally difficult to analyze fully?

Agent-based models typically contain many parameters, causing the parameter space to grow rapidly and making exhaustive exploration computationally prohibitive.

What is the purpose of using machine learning-based surrogates in this study?

Machine learning-based surrogates (metamodels) are used as efficient proxies for agent-based simulations, reducing the cost of exploring parameter combinations.

How do the authors evaluate the classifiers used for categorical metamodeling?

They compare classifiers using indicators of correctness, efficiency, and robustness across different datasets and sample sizes, with an agent-based artificial market as the case study.

Aug 4, 2026

Upload by Oliver