On the feature space and architecture of ABM frameworks

Abstract

Agent Based Modeling (ABM) is a computational paradigm for simulating the actions and interactions of autonomous agents (both individual or collective entities such as organizations or groups) to understand the behavior of a system and the underlying laws governing its outcomes. For the conclusions reached via this technique to be valid, the agent population in the ABM should be representative of – and commensurate with – the population size and rule set under study. Moreover, since the systems under study are complex and hence many of their possible behaviors are unknown, simulations should be executed several times to identify average (and representative) behaviors. From a computational perspective, most research reported in the literature has focused on the impact of performance, scalability and visual representations on the quality of the scientific outcomes obtained with ABM. The present work details an Feature Space Maturity Model (FSMM) which includes significant sections usually overlooked in current ABM frameworks. Advances in Computer Science and the advent of more powerful hardware make it possible to define a reference architecture capable of supporting – at the framework level – robust computational properties, a set of essential features leading to correctness and fidelity to reality. The purpose of this work is manifold. First, it introduces a maturity model for formal assessments of the feature space of ABM and evaluates the five most used ABM frameworks. Second, it proposes a macroscopic analysis of different implementations as a correctness mechanism. Third, it identifies and proposes a reference architecture based on the frameworks reviewed. Finally, it provides a Proof of Concept (POC) implementation based on the reference architecture to evaluate its quality according to the FSMM.