Institut für Computational Science, ETH Zurich
Parameter optimization and sensitivity analysis of an agent-based model in social psychology
In recent years, several computational models have been introduced to the field of social psychology. In this study, we are concerned with the analysis of a theoretical model for self-commitment. This model has been developed by Robert Tobias, EAWAG, within the ATASIS (Application oriented Theory based Architecture for Simulation of Interventions in Social systems) framework. The general model should describe data that has been gathered during a campaign in Santiago de Cuba to reduce deposited solid-waste.Our objective is now to find an optimal model parameter set that describes the data best. In addition, we conduct Parameter Sensitivity analyses of the model, i.e. we study the influence of perturbations in the parameter values onto the model performance. We study both local parameter sensitivities based on partial derivatives, and global parameter sensitivities of the model using the Enhanced Morris' method.In this context, I will also present a novel method for parameter sensitivity identification based on the Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES) optimization procedure. Our method directly uses the information that is acquired during an optimization process to provide local and global parameter sensitivity estimates with no additional evaluations of the cost function.