Solving Challenges at the Interface of Simulation and Big Data Using Machine Learning

Abstract

Modeling & Simulation (M&S) and Machine Learning (ML) have been used separately for decades. They can also straightforwardly be employed in the same study by contrasting the results of a theory-driven M&S model with the most accurate data-driven ML model. In this paper, we propose a paradigm shift from seeing ML and M&S as two independent activities to identifying how their integration can solve challenges that emerge in a big data context. Since several works have already examined this interaction for conceptual modeling or model building (e.g., creating components with ML and embedding them in the M&S model), our analysis is devoted on three relatively under-studied stages: calibrating a simulation model using ML, dealing with the issues of large search space by employing ML for experimentation, and identifying the right visualizations of model output by applying ML to characteristics of the output or actions of the users.