Ecological Model Selection via Evolutionary
Computation and Information Theory
Genetic Programming and Evolvable Machines, 5 (2):
229--241 (Juni 2004)DOI: doi:10.1023/B:GENP.0000023690.71330.42
Zusammenfassung
an evolutionary algorithm-based approach to model
selection and demonstrates its effectiveness in using
the information content of ecological data to choose
the correct model structure. Experiments with a
modified genetic algorithm are described that combine
parsimony with a novel gene regulation mechanism. This
combination creates evolvable switches that implement
functional variable-length genomes in the GA that allow
for simultaneous model selection and parameter fitting.
In effect, the GA orchestrates a competition among a
community of models. Parsimony is implemented via the
Akaike Information Criterion, and gene regulation uses
a modulo function to overload the gene values and
create an evolvable binary switch. The approach is
shown to successfully specify the correct model
structure in experiments with a nested set of
polynomial test models and complex biological
simulation models, even when Gaussian noise is added to
the data.