Towards Explaining Emergent Behavior in Multi-Agent Systems Micro-Parameter Space Structuring with Feature Importance in Heatbug Model

Within this study, we investigate behavior of Heatbug model with feature-based identification and interpretation of regions in parameter space of the model. With a single scenario simulation and predictive modelling of agent target characteristic, we showed possible identification of stable regions in parameter space. Considered parameter space was based on two key characteristics in agent generation, namely environment perceiving error rate (predictive ability) and output heat level (performance). With massive simulation in grid search (3200 simulations with 400 steps and 100 agents) we showed possible interpretation of identified regions in parameter space through various aspects (features) of agent. Finally, we showed identification and interpretation of regions with collective (connected behavior) weakly identified by basic statistics.