Utilização de características zootécnicas e de manejo na pecuária para previsão do peso final e bonificação de bovinos empregando redes neurais artificiais

Abstract

Due to the importance role played by Brazilian beef cattle in national and international markets, the uses of forecast models within this context have been widely used. These models, for the most part, are supplied by a variety of factors, particularly zootechnical conditions of cattle, nutritional aspects, management practices, production costs, among others. Considering this scenario, the main objective of this work is to predict the yield of the cattle of a rural property, through farm weight (PF) and bonus (BN) variables, using the technique of Artificial Neural Networks (ANNs). The methodology used a Feedfoward Backpropagation type network to predict cattle rancher performance using a database of a particular rural property, which contains a series of zootechnical and management variables for a given amount of cattle. Results partially met expected objectives, in which the prediction of the model for the variable farm weight obtained a perfect correlation, good generalization, producing low errors. However, the result for the bonus prediction model presented a very high error, low correlation and poor generalization due to a tool limitation and the choice of data to be used in the network input matrix.