Predicting Weaning Weight of Romanov Lambs From Biometric Measurements Before Weaning Age Using Machine Learning Algorithms.

BACKGROUND

Machine learning systems learn from historical data to forecast future outcomes. In the context of livestock farming, machine learning can be utilized to predict variables such as growth rates, milk production and breeding success by analysing data related to animal health, nutrition and environmental conditions.

OBJECTIVE

This study aimed to investigate the performance of different machine learning algorithms in predicting weaning weight based on biometric measurements of Romanov lambs at 30 days of age.

METHODS

The biometric traits of the lambs, including body length (BL), chest circumference (CC), chest depth (CD), chest width (CH), withers height (WH), rump height (RH), rump width (RW) and sex were used to construct predictive models. The study employed random forest (RF), classification and regression trees (CART), gradient boosting (GB), eXtreme gradient boosting (XGBoost) and CatBoost algorithms. The data was standardized to eliminate scale differences and divided into training (80%) and test (20%) sets. GridSearchCV was utilized for hyperparameter optimization. The performance of the models was evaluated using various goodness-of-fit metrics, including RMSE, MAE, R, MAPE, RAE, MAD and SD ratio.

RESULTS

The gradient boosting and XGBoost models performed the highest R values and the lowest RMSE, MAE and MAPE values in the test data. In contrast, the random forest and CatBoost models showed lower predictive performance, with higher errors in the test data.

CONCLUSION

The study suggests that machine learning algorithms, particularly gradient boosting and XGBoost, show promising potential in predicting the weaning weight of lambs. These insights may facilitate more informed decision-making in animal breeding and selection, potentially contributing to enhanced livestock management practices.