Metamodels for Evaluating, Calibrating and Applying Agent-Based Models: A Review

The development of agent-based modeling is characterized by higher requirements for parameterizing, evaluating, and documenting these computationally expensive models. Accordingly, there is also a growing demand for “easy to use” applications that just mimic the input-output behavior of such models. Metamodels are increasingly being used for these tasks. This paper provides an overview of common types of metamodels and their purposes in the context of agent-based modeling. Implementation efforts and performance were also evaluated to assist modelers in selecting and applying metamodels for their own needs. All applications found were classified according to specific types of metamodels and evaluated by researchers from various disciplines for implementation effort and effectiveness. Specifically, the authors evaluated the performance of the metamodel in accordance with (i) uncertainties, (ii) the assessment of suitability provided by the authors for a specific purpose, and (iii) the number of assessment criteria provided for assessing suitability. 40 different applications of metamodels were selected from studies published in peer-reviewed journals from 2005 to 2019. They were used for sensitivity analysis, calibration and scaling of agent-based models, and to simulate their predictions for various scenarios. This overview provides information on the most applicable types of metamodels for each purpose.