Diploma and Master Theses (authored and supervised):
"Simulation und Validierung von agentenbasierten Modellen für die Ausbreitung von Epidemien";
Supervisor: F. Breitenecker;
Institut für Analysis und Scientific Computing,
final examination: 2013-06-05.
Motivation: Agent-based models are characterized by their detailed view of the system. Thanks
to the development of more powerful computer architectures these type of models are getting more
and more attractive. The aim of this thesis is to describe, validate and simulate an agent- based
model for spreading diseases. Moreover, a calibration on the data of influenza-epidemics in 2006/07
and a simulation for vaccination scenarios should be carried out.
Content: The main part of this thesis is the description of an agent-based model for epidemics.
The introduced model has been developed step-by-step in the last few years. This is why the
modelīs description is changing over time so that the given description should be seen as state of
the art of the current model. When developing a model many aspects as the behaviour of social
systems, individual personal properties in the population, medical processes and the re- alization
of economical examinations have to be observed.
A high grade of model credibility is important for agent-based models of this dimension. If wrong
assumptions are made the model produces wrong results. The process that takes care of this is
called validation. Agent-based models have to deal with high complexity which is why it is
difficult to validate them. There is no standardized validation strategy that can be used. For
every agent-based model individual methods that can be applied should be found. That is why a big
part of this work deals with the research and description of validation methods. The focus is
validation for agent-based models and possible application for the model.
Furthermore, a practical part can be found in this thesis in which the calibration on data from the
influenza-season 2006/07 takes place. For a better understanding the impact of parameters on the
results is shown.
Finally, results of the calibration and vaccination strategies are shown and validation methods are
Results: Under the made assumptions a reproduction of the data of 2006/07 is not possible. If we
doubt the data and calibrate on more infected people, the scaled data can be reproduced by the
model. Based on these fundamental scenarios possible vaccination scenarios are simu- lated and
presented. A calibrated model is able to provide information for decision makers on the
application of vaccines.
Validation methods for agent-based models are found and partly applied. The literature re-
search delivers a wealth of different validation techniques, which can be adapted more or less
well. There is also a strategical approach found to validate agent-based models. In addition a
system called VOMAS can help to validate the model. This system is a monitoring multi-agent system.
It is not possible to validate the medical background information in this thesis. This should be
done by experts of virology, epidemiology or other research institutions. However, some of the
shown methods can be seen as base for their validation.
Conclusion: Due to the failure of calibrating on the original data, it is very likely that the data and
detailed information on the spreading process of the model should be questioned. To find mistakes
a constant validation in cooperation with partners should be conducted during the further
modelling process. This is essential for a correct parameterization of the model. The results
of this work show, that a correct parameterized model can (1) help to understand the infection
process of infectious diseases, (2) help to do economic evaluations, (3) help to predict future
epidemic spreading progress and (4) help to test further vaccination policies in the population.
Electronic version of the publication:
Created from the Publication Database of the Vienna University of Technology.