ScholarWorks@경상국립대학교

초록

Proper management of digestible energy (DE) is crucial for maintaining pig health, promoting growth, and facilitating reproduction by supporting essential biological processes. Therefore, this study sought to predict the digestible energy requirement (DER) in the growing-finishing phase of pigs, where four machine learning (ML) models: multiple linear regression (MLR), support vector regression (SVR), random forest regression (RFR), and multilayer perceptron (MLP) were applied across four datasets, with the input parameters including body weight of pigs (BW), inside temperature (IT), inside relative humidity (IRH), and inside CO2 concentration (ICO2) of pig barns. Two experiments were conducted in 2022 and 2023, involving a total of eighteen 65-day-old crossbred pigs during each experimental period. These pigs were equally divided into three experimental pig barns, each comprising three gilts and three boars, and subjected to different DE content diets. The livestock environment management system sensor and two load cells were used to measure the IT, IRH, ICO2, BW and DE intake of pigs, respectively. The result of the study showed no significant difference in DE intake by pigs with different diets (p > 0.05). While evaluating the model's performance under different datasets, it was observed that the RFR model exhibited the best performance with DS4, explaining over 97 % and 95 % of actual and predicted data during the training and testing phases, respectively, while MLR model demonstrated the worst accuracy (R2 < 0.93 and RMSE >6.1 MJ/pig day) compared to the other models. The ranking of model's performance under DS4 was as follows: RFR > MLP > SVR > MLR. Most importantly, the sensitivity analysis revealed that BW exerted the most significant impact on predicting DER among other variables. In conclusion, the study suggested that RFR model can precisely predict DER, offering valuable insights for pig farmers to enhance their understanding of DE management. © 2025 The Author(s)

키워드