Classification of Carcass Fatness Degree in Finishing Cattle Using Machine Learning

Nowadays, there is an increase in world demand for quality beef. In this way, the Government of the State of Mato Grosso do Sul has created an incentive program (Precoce MS) that stimulates producers to fit into production systems that lead to the slaughter of animals at young ages and superior carcass quality, towards a more sustainable production model. This work aims to build a classification model of carcass fatness degree using machine learning algorithms and to provide the cattle ranchers with indicators that help them to early finishing cattle with better carcass finishing. The dataset from Precoce MS contains twenty-nine different features with categorical and discrete data and size of 1.05 million cattle slaughter records. In the data mining process, the data were cleaned, transformed and reduced in order to extract patterns more efficiently. In the model selection step, the data was divided into five different datasets for performing cross-validation. The training set received 80% of the data and the test set received the other 20%, emphasizing that both had their data stratified respecting the percentage of each target class. The algorithms analyzed and tested in this work were Support Vector Machines, K-Nearest Neighbors, AdaBoost, Multilayer Perceptron, Naive Bayes and Random Forest Classifier. In order to obtain a better classification, the recursive feature elimination and grid search techniques were used in the models with the objective of selecting better characteristics and obtaining better hyperparameters, respectively. The precision, recall and f1 score metrics were applied in the test set to confirm the choice of the model. Finally, analysis of variance ANOVA indicated that there are no significant differences between the models. Therefore, all these classifiers can be used for the construction of a final model without prejudice in the classification performance.