Biomedical Engineering Reference
In-Depth Information
The individuals generated after these cross-
overs could, either be incorporated to an already
existing species or, if they analyse a new area of the
search space, create themselves a new species.
Finally, the GA provides as much solutions as
species remain actives over the search space.
amount (and nature) of other substances added
to the beverages. Particularly, sugar addition is
a common and simple adulteration, though dif-
ficult to characterise. Other adulteration methods,
either alone or combined, involve addition of
water, pulp wash, cheaper juices, colorants, and
other undeclared additives (intended to mimic the
compositional profiles of pure juices) (Saavedra,
García & Barbas, 2000).
Infrared spectrometry (IR) is a fast and con-
venient technique to perform screening studies
in order to assess the quantity of pure juice in
commercial beverages. The interest lies in devel-
oping, from the spectroscopy data, classification
methods that might enable the determination of
the amount of natural juice contained in a given
sample.
However, the information gathered from the
IR analyses has some fuzzy characteristics (ran-
dom noise, unclear chemical assignment, etc.), so
analytical chemists tend to use techniques like
Artificial Neural Networks (ANN) to develop ad-
hoc classification models. Previous studies (Gestal
et al, 2005) showed that ANN classify apple
juice beverages according to the concentration of
natural juice they contained and that ANN had
advantages over classical statistical methods, such
as faster model development and easy application
of the methodology on R&D laboratories. Disap-
pointingly, the large number of variables derived
from IR spectrometry requires too much ANN
training time and, the most important, it makes
very difficult to establish relationships between
these variables and analytical knowledge.
Several approaches were used to reduce this
number of variables to a smaller subset, which
should retain the classification capabilities. In such
way, the ANN training process and the interpreta-
tion of the results will be highly improved.
Furthermore, a previous variable selection
will produce others advantages: cost reduction
(if the classification model requires the use of
a reduced amount of data, the time needed to
PRACTICE
The different proposals provided have been tested
on classic lab problems, as Ackley or Rastrigin
functions, and on real problems as the following
described, related to juice samples. Due to the
complexity of this last problem, the following
subsections will be focus on its description and
on the results achieved.
variable Selection for the
Classification of Apple Beverages
The importance of juice beverages in daily food
habits makes juice authentication an important is-
sue, for example, to avoid fraudulent practices.
A successful classification model can be used
for checking two of the most important corner-
stones of the quality control of juice-based bever-
ages: monitor the amount of juice and monitor the
Figure 6. Overview of the proposed system
Initial Population
(randomly created)
Initial Population
(randomly created)
Initial Population
(randomly created)
Initial Population
(randomly created)
Interspecies
Crossover
Interspecies
Crossover
Definition of
Species
Definition of
Species
Remove Isolated
Individuals
Remove Isolated
Individuals
Goal achieved?
Goal achieved?
No
No
Yes
Yes
End
End
End
End
Search WWH ::
Custom Search