Yang Xidi, Zhu Liangyu, Jiang Wenyu, Yang Yiting, Gan Mailin, Shen Linyuan, Zhu Li
State Key Laboratory of Swine and Poultry Breeding Industry, College of Animal Science and Technology, Sichuan Agricultural University, Chengdu 611130, China.
Key Laboratory of Livestock and Poultry Multi-Omics, Ministry of Agriculture and Rural Affairs, College of Animal Science and Technology, Sichuan Agricultural University, Chengdu 611130, China.
Animals (Basel). 2025 Jun 16;15(12):1773. doi: 10.3390/ani15121773.
Feed conversion ratio (FCR) is a critical indicator of production efficiency in livestock husbandry. Improving FCR is essential for optimizing resource utilization and enhancing productivity. Traditional methods for FCR optimization rely on experience and long-term data collection, which are time-consuming and inefficient. This study explores the feasibility of predicting long-term FCR using short-term FCR data based on machine learning techniques. We employed nineteen machine learning algorithms, including Linear Regression, support vector machines (SVMs), and Gradient Boosting, using historical datasets to train and validate the models. The results show that the Gradient Boosting model demonstrated superior performance, achieving a coefficient of determination (R) of 0.72 and a correlation of 0.85 between predicted and actual values when the testing interval exceeded 40 kg. Therefore, we recommend a minimum feeding measurement interval of 40 kg. Furthermore, when the testing interval was set to 40 kg and further refined to the range of 50-90 kg, the model achieved an R of 0.81 and a correlation of 0.90 for FCR prediction in the 30-105 kg range. Among the 19 machine learning algorithms tested, Gradient Boosting, LightGBM, and CatBoost showed superior performance, with Gradient Boosting achieving the best results. Considering practical production requirements, the 50-90 kg feeding stage proved to be the most ideal for FCR testing. This study provides a more effective method for predicting feed efficiency and offers robust data support for precision livestock farming.
饲料转化率（FCR）是畜牧生产效率的关键指标。提高FCR对于优化资源利用和提高生产力至关重要。传统的FCR优化方法依赖经验和长期数据收集，既耗时又低效。本研究探讨了基于机器学习技术使用短期FCR数据预测长期FCR的可行性。我们采用了19种机器学习算法，包括线性回归、支持向量机（SVM）和梯度提升，利用历史数据集训练和验证模型。结果表明，梯度提升模型表现出卓越的性能，当测试间隔超过40千克时，决定系数（R）达到0.72，预测值与实际值之间的相关性为0.85。因此，我们建议最小喂食测量间隔为40千克。此外，当测试间隔设定为40千克并进一步细化到50 - 90千克范围时，该模型在30 - 105千克范围内FCR预测的R值为0.81，相关性为0.90。在所测试的19种机器学习算法中，梯度提升、LightGBM和CatBoost表现出卓越的性能，其中梯度提升取得了最佳结果。考虑到实际生产需求，50 - 90千克的喂食阶段被证明是FCR测试最理想的阶段。本研究为预测饲料效率提供了一种更有效的方法，并为精准畜牧养殖提供了有力的数据支持。
