An Approach for Face Validity Assessment of Agent-Based Simulation Models Through Outlier Detection with Process Mining

When designing simulations, the objective is to create a representation of a real-world system or process to understand, analyze, predict, or improve its behavior. Typically, the first step in assessing the credibility of a simulation model for its intended purpose involves conducting a face validity check. This entails a subjective assessment by individuals knowledgeable about the system to determine if the model appears plausible. The emerging field of process mining can aid in the face validity assessment process by extracting process models and insights from event logs generated by the system being simulated. Process mining techniques, combined with the visual representation of discovered process models, offer a novel approach for experts to evaluate the validity and behavior of simulation models. In this context, outliers can play a key role in evaluating the face validity of simulation models by drawing attention to unusual behaviors that can either raise doubts about or reinforce the model’s credibility in capturing the full range of behaviors present in the real world. Outliers can provide valuable information that can help identify concerns, prompt improvements, and ultimately enhance the validity of the simulation model. In this paper, we propose an approach that uses process mining techniques to detect outlier behaviors in agent-based simulation models with the aim of utilizing this information for evaluating face validity of simulation models. We illustrate our approach using the Schelling segregation model.