CLASSIFICATION OF DAILY BODY WEIGHT GAINS IN BEEF CATTLE VIA NEURAL NETWORKS AND DECISION TREES

Daily body weight gains are an important indicator for the evaluation of beef production. The objective of this study was to verify the quality and detection abilities of selected data mining methods (artificial neural networks and decision trees) applied to the analysis of daily body weight gains in beef heifers. A total of 261 individuals (Charolais, Charolais x Simmental and Charolais x Hereford) were genotyped and their daily body weight gains were recorded. The genetic analysis included polymorphic variants of leptin, myostatin, and prion protein genes. The animals were divided into two categories: those with daily body weight gains equal to or lower than the mean for the analyzed population (99 heifers) and those with daily body weight gains higher than the mean (112 heifers). Heifer birth weight, breed, birth season, dam body weight at calving, calving interval, and genotypes were used as predictors. From among the analyzed models, the neural network with 8 neurons in a single hidden layer (root mean squared error equal to 0.4471). and the tree with an error coefficient of 0.4076 were selected These classifiers correctly detected 20 out of 27 heifers with daily body weight gains below the mean. In the case of heifers with daily body weight gains above the mean, neural networks correctly classified 17 out of 23 individuals, whereas decision tree only 14. Breed, leptin genotype and combined genotype turned out to be the most important variables contributing to the determination of body weight gains class for both types of classifiers.