Using artificial neural networks to predict pH, ammonia, and volatile fatty acid concentrations in the rumen

摘要

The objectives of this study were (1) to predict ruminalpH and ruminal ammonia and volatile fatty acid(VFA) concentrations by developing artificial neuralnetworks (ANN) using dietary nutrient compositions,dry matter intake, and body weight as input variables;and (2) to compare accuracy and precision of ANNmodel predictions with that of a multiple linear regressionmodel (MLR). Data were collected from 229 publishedpapers with 938 treatment means. The data setwas randomly split into a training data set containing70% of the observations and a test data set with theremaining observations. A series of ANN with a rangeof 1 to 9 artificial neurons in 1 hidden layer were examined,and the best one was selected to compare withthe best-fitted MLR model. The performance of modelpredictions was evaluated by root mean square errors(RMSE) and concordance correlation coefficients (CCC)using cross-evaluations with 100 iterations. When usingthe ANN to predict ruminal pH and concentrations ofammonia, total VFA, acetate, propionate, and butyrate,the RMSE were 4.2, 41.4, 20.9, 22.3, 32.9, and29.7% of observed means, respectively. The RMSE forthe MLR were 4.2, 37.8, 18.3, 19.9, 29.8, and 26.6% ofthe observed means. The CCC for ruminal pH, ruminalconcentrations of ammonia, total VFA, acetate, propionate,and butyrate were 0.57, 0.49, 0.45, 0.40, 0.52, and0.40, using the ANN, and 0.37, 0.48, 0.40, 0.29, 0.43,and 0.35, using the MLR. Evaluations of the MLR andthe ANN indicated that these 2 model forms exhibitedsimilar prediction errors, with 4.2, 39.6, 19.6, 21.1, 31.3,and 28.1% of observed means for pH, ammonia, totalVFA, acetate, propionate, and butyrate. Although theANN increased the precision of predictions related toruminal metabolism, it failed to improve the accuracycompared with the linear regression model.