Classification of Spanish autochthonous bovine breeds. Morphometric study using classical and heuristic techniques

Six morphometric traits (height at withers, height at rump, chest depth, width at hips, width at pins and rump length) were analysed to characterise from a breed point of view 518 females from four autochthonous Andalusian cattle breeds (Berrenda en Colorado, Berrenda en Negro, Cárdena Andaluza and Negra Andaluza). Four methods (one classical and three heuristic) were used to distinguish between the four breeds by morphometric traits: Discriminant Function Analysis (DFA), Multilayer Perceptrons (MLPs) (a type of neural network), Probabilistic Neural Networks (PNNs) and Support Vector Machines (SVMs). Results indicated not only that DFA was overall inferior to the other three methods, but also that it could not be used to distinguish one breed from another when they were genetically very close or related in terms of breeding. MLP and SVM had similar ability to discriminate, both being better than PNN. Sensitivity analysis carried out on the models found to have the best discrimination power indicated that the most important variables were: depth, height at rump and width at pins.