Table 3.
Stability of filter methods for prediction of residual feed intake.
| Subset size1 | Filter method2 | NSNPs3 | PDF4 | NOG5 |
|---|---|---|---|---|
| 50 | mrmr | 500 | 0.30 | 0.53 |
| cforest | 500 | 0.90 | 0.02 | |
| spearcor | 500 | 0.19 | 0.73 | |
| univ.dtree | 500 | 0.96 | 0.00 | |
| random | 500 | 0.98 | 0.00 | |
| 250 | mrmr | 2,500 | 0.26 | 0.58 |
| cforest | 2,500 | 0.80 | 0.04 | |
| spearcor | 2,500 | 0.23 | 0.67 | |
| univ.dtree | 2,500 | 0.85 | 0.01 | |
| random | 2,500 | 0.88 | 0.00 | |
| 500 | mrmr | 5,000 | 0.25 | 0.59 |
| cforest | 5,000 | 0.72 | 0.04 | |
| spearcor | 5,000 | 0.21 | 0.66 | |
| univ.dtree | 5,000 | 0.75 | 0.02 | |
| random | 5,000 | 0.79 | 0.00 | |
| 750 | mrmr | 7,500 | 0.22 | 0.62 |
| cforest | 7,500 | 0.65 | 0.04 | |
| spearcor | 7,500 | 0.20 | 0.68 | |
| univ.dtree | 7,500 | 0.67 | 0.02 | |
| random | 7,500 | 0.71 | 0.00 | |
| 1,000 | mrmr | 10,000 | 0.21 | 0.64 |
| cforest | 10,000 | 0.59 | 0.04 | |
| spearcor | 10,000 | 0.19 | 0.69 | |
| univ.dtree | 10,000 | 0.60 | 0.03 | |
| random | 10,000 | 0.64 | 0.00 | |
| 1,500 | mrmr | 15,000 | 0.19 | 0.66 |
| cforest | 15,000 | 0.49 | 0.04 | |
| spearcor | 15,000 | 0.18 | 0.69 | |
| univ.dtree | 15,000 | 0.50 | 0.03 | |
| random | 15,000 | 0.52 | 0.00 |
1
Subset size = number of selected features.
2
Filter method = Maximum relevance minimum redundancy (mrmr); Random forest (cforest); Spearman’s correlation (spearcor); Univariate decision tree (univ.dtree); Random selection (random).
3
NSNPs = Total number of SNPs selected in the 10 subsets.
4
PDF = Proportion of distinct features in the 10 subsets.
5
NOG = Nogueira et al. (2018) stability estimator.