VisEPS: a visual explorer of parameter spaces for networked models

Abstract

Simulations of complex social systems, such as those represented by epidemiological models, have been very useful in supporting decision makers during the last pandemic. These models generally comprise a high number of parameters, which makes it hard to identify the values that best reproduce the empirical data. Furthermore, different combinations of parameters may achieve a good fit, which renders an automatic solution ill-suited to the task. A human expert is required to make the final decisions about the optimal parameter values. We present VisEPS (Visual Explorer of Parameter Spaces), a framework for visually analyzing the effects of a very large set of parameters, with the aim of fitting a geographically explicit networked model to data obtained during the COVID-19 pandemic. We use a networked extension of a susceptible-infected-recovered (SIR) model to reproduce the epidemic dynamics in the city of Buenos Aires and its neighboring interconnected districts. We overlay binned scatterplots on a map, which facilitates the visual identification of each district and its connections. To further explore the model’s performance against data, additional views, such as parallel coordinates and histograms, along with drill-down mechanisms, have been incorporated. Finally, a use case is described in which the level of connectivity between districts is included in the analysis. The identification of suitable parameter ranges is facilitated by an iterative and incremental process, whereby new sets of simulations are incrementally requested, guided by interactive visual inspections. This permits the exploration of a parameter space that would otherwise be impossible to fully explore.