Inteligência computacional na predição de comportamento, adaptabilidade e estabilidade de algodoeiro de fibra branca e colorida

• Ano de defesa: 2022
• Autor(a) principal: Cardoso, Daniel Bonifácio Oliveira
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Agronomia
• Palavras-chave em Português: Inteligência Computacional; cotton; algodão; Random forest; Computational Intelligence; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::FITOTECNIA::MELHORAMENTO VEGETAL; Agronomia; Algodão - Inteligência computacional - Minas Gerais; Algodão - Cultivo - Minas Gerais; Algodoeiro - Minas Gerais
• Link de acesso: https://repositorio.ufu.br/handle/123456789/36611; https://doi.org/10.14393/ufu.te.2022.5346

To better understand yield variations in crops, in order to predict the behavior of genotypes, the use of computational intelligence can be an alternative, mainly by allowing the insertion of environmental variables in the models. With this in mind, the objective of this work was to evaluate the use of computational intelligence methods for cotton yield prediction using temporal variables in the state of Minas Gerais. The data of productivity are from the municipalities of Minas Gerais with continuous production of cotton in the last ten years: Buritis, Coromandel, Presidente Olegário, São Gonçalo do Abaete and Unaí. The climatological data were obtained month by month, being December, January, February, March, April, May and June, in each year/harvest from 2010 to 2021 being these: maximum temperature (ºC), minimum temperature (ºC), relative humidity (%), soil moisture, moisture in the root zone (%), accumulated precipitation (mm) and average precipitation (mm) totaling 58 variables evaluated. We separated 80% of the input data for model testing and 20% for model training tests, using the Random Forest, Regression Trees and Linear Regression methods in their standard configurations, with the aid of the software Genes and Matlab, as well as Python software. The variables showed significant effect when segmented into months. The machine learning model, based on Random Forest, provides the best prediction results compared to real data, showing promise for estimating productivity in the state of Minas Gerais. The use of computational intelligence for yield prediction using only climatic data can infer with good accuracy to estimate yield data. The machine learning method, Random Forest was the one that obtained the best estimates in all the evaluated cities of Minas Gerais. A more robust database containing more phenotypic information of the cotton plant is needed for better accuracy of the models aiming to predict cotton yields in Minas Gerais.