Aplicação de algoritmos de aprendizado de máquina para classificação da qualidade das carcaças dos lotes de bovinos abatidos: um estudo de caso nos dados do programa Precoce MS

• Ano de defesa: 2023
• Autor(a) principal: Rafael Rodrigues Marquesi
• Orientador(a): Rafael Geraldeli Rossi
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Fundação Universidade Federal de Mato Grosso do Sul
• País: Brasil
• Palavras-chave em Português: Mineração de dados, aprendizado de máquina, algoritmos supervisionados, classificação, pecuária de precisão, agropecuária
• Link de acesso: https://repositorio.ufms.br/handle/123456789/6580

In order to encourage the production of superior quality beef and seeking to meet the criteria of an increasingly demanding market, the Government of Mato Grosso do Sul provides tax incentives, through the Precoce MS program, for producers who slaughter animals with higher quality carcasses and at a young age. The Precoce MS program provides a dataset with information related to the characteristics of cattle, production systems, and the quality of the carcass of the slaughtered animals. However, manually analyzing the data to find factors that may be related to the production of a higher quality carcass may be impractical. In this scenario, Data Mining techniques can be applied to extract useful knowledge and build models to predict carcass quality. Previous works have already applied Data Mining techniques to the Precoce MS program dataset. However, the classification performance was uncertain in current data, state-of-the-art algorithms were not used for tabular data, animals were not used in batches, and other potentially important attributes for predicting the quality of the carcass were not used, such as climatic, nutritional, and commodity price-related attributes. Given this, the present work aimed to use Data Mining techniques, more specifically, algorithms for the construction of classification models to predict the quality of the batch of carcasses, considering: state-of-the-art algorithms, from different paradigms and with different sets of parameters. Furthermore, the most current data from the Precoce-MS program was considered (up to the date of execution of this work) and the data set enriched with meteorological, nutritional and commodity pricing attributes. The results showed that the Random Forest Classifier algorithm presented the best classification performance (72.63% Accuracy). Finally, using the best model, a REST API was developed to classify the batch of cattle to be slaughtered.