Please use this identifier to cite or link to this item: http://repositorio.ufla.br/jspui/handle/1/33256
Title: Redes bayesianas na predição de valores energéticos de alimentos para aves
Other Titles: Bayesian networks in the prediction of energy values of feedstuffs for poultry
Authors: Lima, Renato Ribeiro de
Bueno Filho, Júlio Sílvio de Sousa
Muniz, Joel Augusto
Rodrigues, Paulo Borges
Mariano, Flávia Cristina Martins Queiroz
Keywords: Algoritmo híbrido mmhc
Distribuição empírica
Energia metabolizável
Equações de predição
Nutrição de aves
Empirical distribution
Hybrid algorithm mmhc
Metabolizable energy
Nutrition of monogastric
Prediction equations
Issue Date: 21-Mar-2019
Publisher: Universidade Federal de Lavras
Citation: ALVARENGA, T. C. Redes bayesianas na predição de valores energéticos de alimentos para aves. 2019. 101 p. Tese (Doutorado em Estatística e Experimentação Agropecuária)-Universidade Federal de Lavras, Lavras, 2019.
Abstract: Balanced diets for poultry nutrition depend on the knowledge of the chemical composition of the feedstuffs, especially the values of apparent metabolizable energy corrected by the nitrogen balance (EMAn). The values of EMAn can be obtained in biological assays, in which the execution is time-consuming and expensive, as well as by feedstuff composition tables. Another way of obtaining the values of EMAn are the prediction equations established according to the chemical composition of the feedstuffs, usually of easy and fast obtaining. In the literature there are studies that obtained the prediction equations through multiple regression, meta-analysis and neural networks. In order to find more accurate results, the Bayesian networks are used to predict EMAn according to the chemical composition of the feedstuffs. Bayesian networks are graphical models (graphical models), which consist of graphical (graph) and probabilistic representation (conditional and joint probability distributions) of the variables. Bayesian networks were proposed by Judea Pearl, then known for defending probabilistic knowledge in the field of artificial intelligence. For a broad understanding of this area of research, Thompson Reuters’ Web of Science database was used to identify the patterns and trends of scientific publications on Bayesian networks, thus making it possible to check that most publications are related to the area of Computer Science. In the applied areas, mainly agriculture and livestock, there are still very few publications, however, Bayesian networks is an unprecedented research line in poultry nutrition and can be studied by researchers who are interested in predicting the values of metabolizable energy. Equations have been proposed through of the Bayesian networks, their being obtained by the Max-Min Hill Climbing algorithm (MMHC) and they can be used by the broiler industry in the making of diets, since they presented accuracy in the prediction of EMAn. Moreover, they were validated with data from metabolic assays and showed both precision and accuracy in the prediction of energy values. These equations are available in a calculator that can be installed on phones, tablets and computers. In this thesis a new methodological approach was also proposed in which it considered uncertainties in obtaining equations from the results of hybrid Bayesian networks. The estimates from means and mode of the coefficients of the chemical composition of feedstuffs derived from empirical distributions constructed with 10000 hybrid Bayesian networks performed better. The proposed equations showed accurate results as they were evaluated with metabolic assay data. In short, it has contributed both in methodological terms and in practical terms and the production of innovative technological products in agricultural experimentation, more specifically in poultry nutrition.
Description: Arquivo retido, a pedido da autora, até março 2021.
URI: http://repositorio.ufla.br/jspui/handle/1/33256
Appears in Collections:DES - Estatística e Experimentação Agropecuária - Doutorado (Teses)

Files in This Item:
There are no files associated with this item.


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.