SimAnalyzer: automated description of groups dynamics in agent-based simulations

Multi-agents based simulations (MABS) have been successfully used to model complex systems in multiple areas. However, a pitfall of MABS is that their complexity increases with the number of agents and behaviors considered in the model. For average and large systems, different phenomena can simultaneously occur at different intermediate levels and influence each other [2]. For instance, groups of agents (flocks of birds, social groups, etc.) following similar state’s trajectories may appear, evolve and disappear. To describe and evaluate the evolution of groups, the observation of global and individual variables (like in [1]) is not sufficient anymore. Moreover, because of the emergent properties of complex systems, those groups may be unexpected, or their presence may even be unnoticed because no suited variable or any other adapted observation mechanism is provided in the simulator. The significance and even the existence of groups can then be hidden by the usually huge amount of available data. In this paper we introduce the use of statistical based tools to assist the modeler in discovering, describing and following the evolution of groups of agents, by combining data clustering and value test. Our model can be described within 5 main steps which will be illustrated with a NetLogo library model example.