Fixing Equifinality of ABM: sequential parameter space searching method based on global sensitivity analysis

Keywords: ABMs, Agent-Based model, GSA, equifinality, Infectious disease, Contact
Abstract Type: Paper Abstract

Abstract

Various studies have been conducted in many fields to curb infectious disease. Among them, application of ABMs (Agent-Based Model Simulation) has been built to illustrate multiple infectious situations and to find out relationship between spatiotemporal parameters and the number of infections. Since numerous different situations are intertwined in infections, however, we need improved simulation calibrations and validation to make model fit reality.

Equifinality, which stands for the fact that many simulation models could match with origin data equally well would cause the acceptance of an invalid model. Many ABMs application studies in geography field have used single observed data to calibrate multiple parameters by assuming the model fits reality if the simulation outcomes match well with the pattern of observed data. However, the more parameters we use to delineate simulation in detail to reflect the fine scale of reality, the larger number of combinations and ranges of parameter value that could match up with observed pattern equally well, which cause uncertainties of interpretating interactions between parameters and the outcomes.

To address this issue, we suggest ‘multivariate outcome-based GSA (Global Sensitivity Analysis) with calibration’ that gradually reduce the range of parameters in ABMs with multiple observed data. For this, we build a pedestrian ABMs to illustrate movement in indoor space. For observed data, we use an indoor trajectory data, which captures the movement of people in an indoor conference venue. Next, we investigate which parameters have a great impact on the number of contacts in space.