Prediction of carcass tissues and primal cuts of goat kids through machine learning based on carcass image analysis

Knowledge of carcass composition enables its use in animal selection, feed management, and the development of objective carcass classification systems. In the autochthonous goat breeds production, there is limited information on carcass composition and the availability of methods capable of predicting it. The objective of this study was to predict the tissue composition and carcass cuts of goat kids through machine learning based on carcass image analysis. Video image analysis (VIA) carcass measurements and hot carcass weight (HCW) were used to predict the weights of muscle, subcutaneous and intermuscular fat and bone tissues in goat kid carcasses using artificial neural network (ANN) and support vector machine (SVM) models. The ANN models generally achieved higher prediction accuracies for tissues (R²=0.91–0.97) and cuts (R²=0.96–0.99) than the SVM models (R²=0.52–0.97 for tissues; R²=0.74–0.98 for cuts). The SVM models generally demonstrated a lower root mean square error (RMSE) value. These results indicate that it is possible to use the VIA measurements and HCW with the ANN and SVM models to predict the quantity of goat kid carcass cuts and tissue composition, with the model selection being an important factor in prediction performance.