Applying statistical debugging for enhanced trace validation of agent-based models. (April 2017)

• Record Type: Journal Article
• Authors: Gore, Ross J; Lynch, Christopher J; Kavak, Hamdi
• Is Part Of: Simulation. Volume 93:Number 4(2017)
• Page Start: 273
• Page End: 284
• Publication Date: 2017-04
• Subjects: Statistical debugging -- trace validation -- agent-based models -- emergent behavior; Computer simulation -- Periodicals; 003.3
• DOI: 10.1177/0037549716659707
• Languages: English
• ISSNs: 0037-5497

Abstract

The verification and validation (V&V) of agent-based models (ABMs) is challenging. The underlying structure of the model and the agents can change over time. Furthermore, the theoretical context of the model is often very different from established models of the same phenomenon. In an effort to overcome these issues, trace validation is becoming a common V&V mechanism within the agent-based modeling community. In trace validation, characteristics of agents and the model are tracked over time and then analyzed by subject matter experts (SMEs) to gain insight into unexpected and potentially invalid output. Here, we present our tool, the V&V Calculator, which applies predicates employed in the field of software engineering. The result is a structured trace validation approach with quantifiable measures that facilitates SME exploration and insight into the causes of unexpected output within ABMs.