Agriculture Reference
In-Depth Information
• Inventory control.
• Linear programming.
• Non-linear programming.
• Goal programming.
• Integer programming.
• Dynamic programming.
• Time series analysis.
• Queuing analysis.
• Risk analysis.
• Simulation.
• Neural networks.
• Genetic algorithms.
• Fuzzy logic theory.
• Chaos theory.
• Game theory.
• Data mining.
precision and lack of prediction bias of
these algorithms.
The theory and other applications of
neural networks is discussed in another
chapter of this topic (see Ferraudo, Chapter
7,
this volume). However, this author suggests
that this tool should be integrated into
traditional biological modelling to advance
further their application to commercial con-
ditions of poultry production where mech-
anistic or empirical models have failed to
provide the accuracy of predictions de-
manded by the industry. Nowadays, many
statistical software packages offer this tool
and the algorithms could be integrated into
existing poultry modelling software, mak-
ing it easier to apply with the right set of data
(Roush, 2006).
The reader will notice that meta-analyses,
frequently used to summarize published in-
formation in poultry science, were not in-
cluded in this list. Although meta-analyses
can be used to explore the tendencies of bio-
logical responses when appropriate statis-
tical methods are used, most of the time the
sources of data are extremely variable, do
not have similar treatment levels and are af-
fected by many unknown variables, which
may not be mentioned in the publications.
This problem in statistics is often called
study heterogeneity (Nordmann
et al
., 2012).
Ideally, the studies whose results are being
combined in the meta-analysis should all be
undertaken in the same way and to the same
experimental protocols (Sutton
et al
., 2000),
but this hardly ever occurs in the meta-
analyses conducted with poultry data. In al-
most all meta-analyses, this inconsistency
among experimental protocols may make the
results of such mathematical exercises in-
accurate and misleading, and the resultant
equations should probably not be used as
simulation, prediction or optimization tools.
Meta-analyses are still valuable as they help
researchers to better understand the factors
that may cause variability in the responses ob-
served when similar treatments are applied,
but they always should lead to controlled
experiments and mechanistic models. Con-
sequently, meta-analyses themselves should
not be considered directly as a modelling
methodology.
Several of these techniques could be ap-
plied to common problems in poultry
production. This includes areas such
as resource distribution and allocation,
planning and orientation, scheduling and
routing, market forecasting, inventory con-
trol, optimization and replacement and
maintenance.
Among the techniques listed, artificial
neural networks are interesting mathemat-
ical tools that have been used in poultry
production in many areas. There are sev-
eral publications that reference the multi-
plicity of applications of this methodology,
which is an output of artificial intelligence
technology. Artificial neural networks have
been evaluated to predict biological re-
sponses of poultry, such as growth (Roush
et al
., 2006; Ahmad, 2009; Ahmadi and
Golian, 2010; Mottaghitalab
et al
., 2010),
egg production in layers (Savegnago
et al
.,
2011; Wang
et al
., 2012), or in breeders
(Salle
et al
., 2003; Faridi and Golian, 2011)
and hatchability (Bolzan
et al
., 2008). Arti-
ficial neural networks have also been ap-
plied to estimate nutrient composition in
feed (Cravener and Roush, 2001; Perai
et al
., 2010; Ebadi
et al
., 2011) and manure
(Chen
et al
., 2009), traits of carcasses (Fari-
di
et al
., 2012) or eggs (Patel
et al
., 1998),
and forecast market behaviours (Huang
et al
., 2009), among other applications. It is
known that neural networks do not help to
understand the systems under study, but it
is impossible to deny the high accuracy,
