Mechanically Deboned Poultry Meat (MDPM) is constituted of the neck and back from chicken carcasses that are extracted in machine. An artificial neural network of the Back-Propagation type was built to determine the amount of MDPM in the composition of commercial foods. The objective of this work was to train, evaluate and apply a network of the Back-Propagation type, with three layers of neurons, in predicting the amount of MDPM in relation to the amount of minerals in the sausage. We used the mineral composition of 29 product samples that contained different amounts of MDPM and 23 commercial samples. The topology of the network was a 5-5-1. The average quadratic error in the training group was of 2.4%, while in the test phase it was 3.8%. The application of the network to commercial samples was inadequate due to diversity of the ingredients used in the training and those found in commercial samples. The neural network developed for determining the MDPM was shown to be efficient in both the training phase and the test.
artificial neural network; backpropagation; mechanically deboned poultry meat
