Machine learning algorithms for the prediction of EUROP classification grade and carcass weight, using 3-dimensional measurements of beef carcasses

Abstract

IntroductionMechanical grading can be used to objectively classify beef carcasses. Despite its many benefits, it is scarcely used within the beef industry, often due to infrastructure and equipment costs. As technology progresses, systems become more physically compact, and data storage and processing methods are becoming more advanced. Purpose-built imaging systems can calculate 3-dimensional measurements of beef carcasses, which can be used for objective grading.MethodsThis study explored the use of machine learning techniques (random forests and artificial neural networks) and their ability to predict carcass conformation class, fat class and cold carcass weight, using both 3-dimensional measurements (widths, lengths, and volumes) of beef carcasses, extracted using imaging technology, and fixed effects (kill date, breed type and sex). Cold carcass weight was also included as a fixed effect for prediction of conformation and fat classes. ResultsIncluding the dimensional measurements improved prediction accuracies across traits and techniques compared to that of results from models built excluding the 3D measurements. Model validation of random forests resulted in moderate-high accuracies for cold carcass weight (R2 = 0.72), conformation class (71% correctly classified), and fat class (55% correctly classified). Similar accuracies were seen for the validation of the artificial neural networks, which resulted in high accuracies for cold carcass weight (R2 = 0.68) and conformation class (71%), and moderate for fat class (57%).DiscussionThis study demonstrates the potential for 3D imaging technology requiring limited infrastructure, along with machine learning techniques, to predict key carcass traits in the beef industry.