REPOSITORY > RESULTS

Abstract

Simulation models are a widely used tool for modelling and simulating systems for which
it is hard to obtain real data. However, the simulation models are usually complex and
it is not an easy task to induce new knowledge and find relationships and dependencies
among different parts (parameters, processes, modules) of the simulation model.
Previous attempts to analyze the outputs from simulation models were using mostly
statistical methods and neural networks, where the main goal was to speed up the simu-
lation process, or to improve the parametrization of the simulation models. In this thesis
we are proposing a methodology for analyzing results of complex simulation models. The
methodology combines simulation outputs, background knowledge, and machine learning,
to obtain new and interesting knowledge about a certain problem of interest.
We apply our methodology to three different simulation models that simulate the co-
existence between genetically-modified and conventional crops at different levels. The
induced machine learning models provide us with new co-existence knowledge about the
positive and negative influences on the co-existence between genetically-modified and
conventional crops. The results encourage us to try the same methodology on different
types of simulation models and different scientific areas. They also pose other challenges
for development of new machine learning methods.