Biomedical Engineering Reference
In-Depth Information
variables related with low concentrations and
with high concentrations.
Best results were obtained using a multimodal
GA, as it was expected, due to its ability to maintain
the genetic individuals homogeneously distributed
over the search space. Such diversity not only
induces the appearance of optimal solutions, but
also avoids the search to stop on a local minimum.
This option does not provide only a solution but
a group of them that have similar fitness. This
allows Chemistry scientists to select a solution
with a sound chemical background and extract
additional information.
FUTURE TRENDS
Due to the fact that ANN are sometimes exces-
sively used, the quality of the variable selection
is quite emphasised, although the time required
for performing such selection is not a critical
factor. As the higher computational load lies in
the training of the inner ANN, several trainings
could be performed simultaneously by means of
the proposed distributed execution.
The proposed system also uses a back propa-
gation learning based-ANN for evaluating the
samples. Other variable selection models, arisen
from other type of networks (LVQ, SOM, etc.),
are being developed for performing a more com-
plete study.
REFERENCES
Ballester, P.J., & Carter, J.N. (2003). Real-Pa-
rameter Genetic Algorithm for Finding Multiple
Optimal Solutions in Multimodel Optimization.
Proceedings of Genetic and Evolutionary Com-
putation (pp. 706-717)
Darwin, C. (1859). On the Origin of Species . John
Murray, London.
FUTURE RESEARCH DIRECTIONS
Deb, K. (2003). A Population-Based Algorithm-
Generator for Real-Parameter Optimization.
(Tech. Rep. No. KanGAL 2003003)
Most of the effort would be focused on the devel-
opment and test of new fitness functions. These
functions would be able to assess the quality of a
sample and would try to require less computational
requirements than the ANN actually used.
Another option to preserve the ANN use
would consist on the distribution of the genetic
individual's evaluation. It would be interesting to
compare both approximations.
Deb, K., & Agrawal, S. (1995). Simulated binary
crossover for continuous search space. Complex
Systems 9(2) (pp. 115-148)
DeJong, K.A., (1975). An Analysis of the Behav-
iour of a Class of Genetic Adaptive Systems. Phd.
Thesis, University of Michigan, Ann Arbor.
Eshelman, L.J., & Schaffer, J.D. (1993). Real coded
genetic algorithms and interval schemata. Foun-
dations of Genetic Algorithms (2) (pp. 187-202)
CONCLUSION
Several conclusions can be drawn from the dif-
ferent proposals for variable selection:
First of all, the classification results were, at
least, acceptable, so different techniques based on
the combination of GA and ANN demonstrated
that they are valid approaches and can be used
to perform variable selection.
Freeman, J. A., & Skapura, D.M. (1991). Neural
Networks: Algorithms, Applications, and Pro-
gramming Techniques , Addison-Wesley, Read-
ing, MA.
Gestal, M., Gómez-Carracedo, M.P., Andrade,
J.M., Dorado, J., Fernández, E., Prada , D. &
Search WWH ::




Custom Search