Desenvolvimento de um método de classificação da conformação de carcaças de ovinos baseada no K-Nearest Neighbor (KNN).

Abstract

The selection of meat production animals has the objective of visualizing the set of characteristics responsible for increasing the economic efficiency of the farm in order to meet consumer expectations. Following this principle, the value of a given carcass, that is, its acceptability to a series of market requirements, depends on the quantity of the quality of the edible portion. The evaluation of the conformation, which expresses the muscle profiles, is a tool used to classify carcasses in terms of commercial value. In this context, the use of computational tools for data processing in search of patterns can optimize this type of evaluation. The objective of this study is to apply a methodology that has the capacity to classify ovine carcasses in terms of their conformation based on attributes such as weight, morphometric and area of the loin-eye area (LEA) of carcasses using the k-Nearest Neighbor (KNN) smart classifier. A total of 72 carcasses of castrated lambs kept under a feedlot system belonging to commercial herds located in the state of Pará were evaluated. A total of 29 morphometric measurements were taken on whole cold carcasses, in addition to weight and LEA. These were classified according to the weight in light and heavy and, later, according to their conformation adopting the SEUROP system by an evaluator by means of digital images of the cold carcass. The execution of the KNN algorithm followed the leave-one-out cross validation methodology, where different learning strategies were used to predict the conformation of the carcasses, whose training sets were distinguished in number of examples (carcasses), attributes (variables ).In the numerical sets, PECC of 61.11% (k = 7) was obtained for strategy 1; 62.50% (k = 5) for strategy 2; 50.94% (k = 5 and 23) for strategy 3 and 58.49% (k = 11) for strategy 4. In the nominal sets referring to Strategies 1, 2, 3 and 4 (k = 1) %. The KNN algorithm was able to predict the conformation scores from morphometric measurements. From the results of accuracy, class sensitivity and classification accuracy, the best strategy was strategy 2 (CONF_72_CP), with attribute reduction, able to optimize the prediction of classification in relation to the other strategies.