Abstract for: Exploring Loop and Parameters Influence using Sensitivity Analysis with Explanatory Model Analysis

Sensitivity analysis can be used to identify influential model parameters, for example, which parameters might be associated with overall payoff values that capture a model’s “success criteria”. Our approach, similar in theme to statistical screening, generates a sensitivity dataset from a system dynamics model, and explores two aspects in order to provide model insights: (1) the influence of parameters and (2) the influence of loop activation/deactivation. The sensitivity data is used to train a supervised learning regression model (random forest). The explanatory analysis methods used are adopted from the domain of statistical/machine learning, where the goal is to generate explainable insights from statistical models. The sensitivity data can be generated with any SD tool, and all the additional analysis is performed using the R tidymodels metapackage.