Neural Networks and Artificial Intelligence in Stockbreeding and Forecasting Dairy Cattle Productivity

Abstract

AbstractStockbreeding in the dairy farm is carried out by highly qualified specialists. To increase milk yields and milk fat content, during the selection of pairs, the specialists consider the information about their ancestors—mothers, mothers of mothers, and mothers of fathers. Based on the analysis of information (milk yield and fat content) on the first lactation of the mother, the mother of the mother, and the mother of the father, the authors create a neural network prediction of the daughter’s milk yield and fat content. The experiment used a forward propagation neural network with two hidden layers of 18 and 10 neurons each. Various productive properties of ancestral combinations of the mother, the mother of the mother, and the mother of the father were used as input streams. The training sample consisted of 200 animals with information on their milk yields and fat content in the first lactation. The test sample included 20 animals. The test results show that the neural network is quite good at predicting productivity results. However, the analyzed information is insufficient to increase the prediction accuracy even with the increased number of ancestors. Presumably, it is necessary to consider not only the milk yield and fat content of the ancestors but also the quantity and quality of feed, the state of the veterinary service, the second and subsequent lactations, and other zootechnical information. It is also necessary to conduct additional experiments with the neural network architecture to increase the volume of the training sample.KeywordsNeural networkMultilayer perceptronStockbreedingArtificial intelligenceDairy farmingPairingJEL CodesC10C45Q16