Calibration of a stochastic agent-based model for re-hospitalization numbers of psychiatric patients

Calibration is a vital part of the modeling and simulation process and denotes the determination of parameter values by estimating them from comparison between simulation results with reference data. During recent decades, a lot of algorithms have been developed for this purpose which are able to fit mentioned parameter values generically without information about the modelled system or the simulation. Especially for stochastic simulation models these routines very often require thousands of iterative simulation executions which, in case of large agent-based models, might be too time intensive. In this paper, we illustrate a real-world example for such a problem and present a solution for it based on probability theory. Hereby we not only calibrate the desired parameters, but also find a measure for the quality of the fit as well. By presenting this example we want to motivate modelers to analyze agent-based models to save costly calibration time.