Prediction of dark cutting carcasses in cattle using machine learning algorithms with stockperson actions and animal behaviors at abattoir: A study in Türkiye

Abstract

The aim was to investigate the relationship between stockperson actions, animal behaviors at the abattoir, and the occurrence of dark cutting in cattle using various machine learning (ML) algorithms. Season, age, sex, breed, carcass bruising score, carcass weight, and various transportation-related variables were also considered as covariates and potential predictors of dark cutting. Data was collected from 648 cattle, including Holstein, Brown Swiss, and Simmental breeds. The percentage of dark cutting carcasses was 6.64 %. The synthetic minority oversampling technique (SMOTE) was used to transform unbalanced dataset into balanced one. ML was applied with four different models, defined based on the inclusion of covariates, stockperson actions, and animal behaviors as predictors. The highest accuracy value (0.97) was obtained with Boosting algorithm. In all algorithms, the highest accuracy values were achieved with models that included stockperson actions as predictors. Age, prod use and beating at slaughter corridor, and lairage type were most important features influencing dark cutting according to Boosting algorithms. In conclusion, the classification of normal and dark cutting carcasses can be achieved with a satisfactory accuracy using the Boosting and Random Forest algorithms with the model including stockperson actions, animal behaviors and various covariates. However, this study reflects local cattle handling practices in Türkiye; further studies are needed to explore cattle handling practices in other countries.