Application of the Different Machine Learning Algorithms to Predict Dry Matter Intake in Feedlot Cattle

English (OpenAlex)

Due to the development of computing technology and different machine learning models, big data sets have gained importance in animal science as well as in many disciplines. The main objective of this study was to compare different machine learning algorithms to predict daily dry matter intake (DMI) in feedlot cattle. The data consisted of 2660 cattle pens placed on feed between January 1988 and December 1997. Machine learning methods were compared in heifers and steers, with 718 in pens of heifers and 1942 in pens of steers. Initial body weight, days on feed, and average proportion of dietary concentrate were used as independent variables to predict DMI in steers and heifers separately. The multivariate linear regression (LR), random forest (RF), gradient boosting regressor (GBR), and light gradient boosting machine (LGBR) algorithms were compared in terms of several performance metrics (MAE, MAPE, MSE, and RMSE). Results showed that the determination coefficient alone is not a good single criterion. It is recommended that the interpretation of model consistency should also consider MAE, MAPE, MSE, and RMSE values. In the current study, all machine learning algorithms yielded similar and lower performance metrics. However, the LGBR and GBR algorithms, were found to perform slightly better than the other algorithms, especially in heifers. Increasing the number of animals and using different independent variables that are related to the DMI can affect the accuracy of DMI prediction.