Artificial intelligence for the prediction of the beef quality before slaughtering the cattle

In this work we aim to carry out assessments on the quality of beef before slaughter by machine learning methodologies. We evaluate the possibility of classifying meat quality based on a limited number of easily detectable quantitative and qualitative characteristics collected before slaughtering the cattle and applying machine learning methodologies on them. Namely, we develop some models which are able to determine 4 qualitative classes (output) of meat ranging from poor to excellent, plus two intermediate ones. To experiment with methodologies we trained them on data collected through a small entrepreneurial reality located in Amorosi, southern Italy, which deals with cattle breeding in stables. Using a simple artificial neural network the classification error percentage stands at 4.32% while the correlation between the predictions and the real outcomes is greater than 0.95. The extension to a further set of machine learning methodologies on a widely used, open-source machine learning platform make it possible to improve the results, allowing to achieve the perfect classification of the meat. Namely Random Forest reaches 0% error percentage. Overall, the study evidences the potential of Machine Learning methods for predicting beef quality before slaughtering the cattle, and provides a framework for incorporating a wide range of factors into such models. This implies the potential to improve the average quality of the product, being able to intervene on breeding procedures, benefiting both producers and consumers.