Utilização de modelos preditivos para avaliação do desempenho produtivo de porcas

Abstract

This study aims to evaluate predictive models on the productive performance of first and second farrowing sows as tools for decision-making in selecting and discarding females in the herd. In the first study, sows were classified based on totals born alive (TBA). In the second study, confusion matrices and decision trees approaches were applied. Productive information of 8869 sows from first to eighth parity was used, considering the variables: number of weaned piglets (Desm), gestation duration (GL), farrowing duration (FD), number of artificial inseminations (AI), weaning to estrus interval (WSI), proportion of dead at weaning (DW), proportion of dead at farrowing (DF), total culling of sows (Descarte). A TBA as predictors in the models. Study 1: three classes were elaborated (high – H: >16 TBA, medium – M: between 12 and 16 TBA, and low – L: <12 TBA) according to the productive performance at the first and second births. Study 2: the data were divided into two groups: training and compared to elaborate the confusion matrices and decision trees. After classifying the information, verifying its biological coherence, correlations, ANOVA tests, and LSD mean comparison tests were performed at a 5% significance level. The combination of productive information from the first two parities can efficiently predict the longevity and productivity of sows. The method can identify the productive performance of superior and inferior sows, helping in decision-making for selection or disposal. The decision trees were able to accurately present the relationships and patterns between production variables and disposal. Regarding productivity, sows that presented TBA above 14 piglets were classified as good and above 16 as excellent. Sows with high and low yields spend less time in the herd. From the aspect of culling criteria, sows with weaning to estrus interval 3.8 to 4.3 days, were discarded earlier, and sows that wean between 13 and 14 piglets remain longer in the herd. The elaboration of classes according to productivity and the construction of decision trees are efficient and accurate methods to elaborate sow selection and discard programs based on the performance data of the first two parity orders.