Table 1.
Prediction performance of the animal-based (ANIM-B) models (CH4 production; g/d) developed using conventional method and four machine learning methods.
| Conventional | glmmLasso | LASSO | SCAD | RF-B | |
|---|---|---|---|---|---|
| RMSPE | 2.96 | 2.85 | 2.85 | 3.00 | 2.91 |
| Reduction of RMSPE (%) | – | 3.80 | 3.62 | -1.25 | 1.52 |
| MAE | 2.29 | 2.18 | 2.17 | 2.29 | 2.21 |
| Reduction of MAE (%) | – | 4.60 | 5.08 | -0.39 | 3.50 |
| CCC | 0.64 | 0.70 | 0.71 | 0.66 | 0.68 |
| Increase of CCC (%) | – | 9.49 | 9.80 | 2.64 | 5.29 |
*The conventional method only used animal-related data; the relative abundance of all the microbial data was log-transformed; glmmLasso, generalized linear mixed model combined with LASSO; LASSO, least absolute shrinkage and selection operator; SCAD, smoothly clipped absolute deviation implemented on linear mixed models; RF-B, random forest combined with boosting. The data were randomly split into a training set and a testing set (80:20) 200 times and were standardized by mean centering and scaling (detailed in “Methods”).