Abstract
Methane (CH(4)) produced by methanogenic archaea during the rumen fermentation of feed carbohydrates leads to global warming and total energy loss. This study aims to compare the accuracy of multiple linear regression (MLR) models and backpropagation neural network (BPNN) in predicting ruminal CH(4) production of the carbohydrate (Carbs) components of the Cornell Net Carbohydrate and Protein System (CNCPS) in mixed rations of beef cattle with different concentrate-to-forage (C/F) ratios. Two datasets were established using the in vitro fermentation method of Menke and Steingass. One of the datasets contained 60 mixed rations with C/F ratios of 30:70, 40:60, 50:50, 60:40, 70:30, 80:20, and 90:10, respectively, which were used to develop CH(4) prediction models. Another dataset included 10 mixed rations with the same C/F ratios, which were used to validate and compare the accuracy of the prediction models. Results indicated that there was a significant multiple regression relationship between CH(4) production and the Carbs-components (CA (sugars), CB(1) (starch and pectin), CB(2) (available cell wall), CC (unavailable cell wall)) of CNCPS (r(2) = 0.91, p < 0.0001). An optimal BPNN model with 2 hidden-layer neuron nodes was established with the same variables (r(2) = 0.93, p < 0.0001). The findings demonstrated that both MLR and BPNN models (p < 0.0001) were suitable for predicting CH(4) production using the Carbs components (CA, CB(1), CB(2), CC) of CNCPS. However, compared with the MLR model, the BPNN model has a greater coefficient of determination (r(2)) value and concordance correlation coefficient (CCC), and a lower root mean square prediction error (RMSPE), demonstrating better prediction performance.