Abordagem neural para a determinação de dependências entre variáveis do processo de transporte e índice de contusões em bovinos

Abstract

The demands on the quality of beef by consumers are increasing. Depending on how they occur, the procedure of transporting cattle to slaughter influence the quality of meat produced by the slaughterhouse and sent to the final consumer. People have been concerned not only with the texture and softness of meat, but also with animal welfare. During transport, from embarkment to landing, the number of contusions associated with the transportation process results in financial losses for producers and slaughterhouses, as well as important questions regarding ethics in dealing with animals. In order to improve the animal welfare and the quality of the animals being transported, this work aims to develop a neural approach, aiming to determine the dependencies between the variables when transporting cattle. The data evaluated were collected at Frigorífico Silva, Santa Maria (RS), for a period of one year, by researchers from the Federal University of Santa Maria. These data are separated by lot, with the amount of 3283 records. The neural network used was the multilayer textit Perceptron, using the algorithm for training error backpropagation. The obtained accuracy was 94,5% using four hidden layers, for activation of these layers was used the function Elu, and for the output layer the function textit softmax.