Feat | Info | Cl | Perf forests | Perf jungles | Imp |
---|
10 | 3 | 2 | 0.85 (0.02) | 0.85 (0.02) | 0.0% |
20 | 10 | 3 | 0.83 (0.02) | 0.84 (0.02) | 1.2% |
50 | 40 | 4 | 0.64 (0.03) | 0.69 (0.03) | 7.8% |
100 | 90 | 5 | 0.49 (0.03) | 0.62 (0.03) | 26.5% |
200 | 150 | 6 | 0.3 (0.03) | 0.45 (0.03) | 50.0% |
300 | 270 | 7 | 0.22 (0.03) | 0.35 (0.03) | 59.1% |
400 | 350 | 8 | 0.18 (0.03) | 0.29 (0.03) | 61.1% |
500 | 400 | 9 | 0.14 (0.02) | 0.22 (0.03) | 57.1% |
1000 | 500 | 10 | 0.11 (0.02) | 0.15 (0.03) | 36.4% |
1000 | 800 | 10 | 0.12 (0.02) | 0.17 (0.03) | 41.7% |
- Each line shows the results of 30 replicate experiments, with 5-fold cross validation, 100 independent runs per fold, forests of size 100, and resultant jungles of size 10,000. Feat: number of features in the dataset. Info: number of informative features. Cl: number of target classes. Perf forests: mean performance of forests on test set across all replicates (SD). Perf jungles: mean performance of jungles on test set across all replicates (SD); a jungle’s output was computed through straightforward majority voting. Imp: Percent improvement of Perf jungles vs. Perf forests