Table 5.
Classification/prediction error (SD), p = 500
| Scenario 1 | Scenario 2 | Scenario 3 | Scenario 4 | |||
|---|---|---|---|---|---|---|
| RF | 23.5% (3.4%) | 9.44 (1.37) | 9.13 (0.59) | 6.74 (0.77) | ||
|
|
22.0% (3.6%) | 7.88 (1.44) | 8.02 (0.81) | 5.08 (0.60) | ||
| RF-log(p) | 18.5% (4.2%) | 5.34 (1.69) | 8.65 (1.48) | 3.57 (0.43) | ||
| ET | 20.9% (4.0%) | 5.92 (1.61) | 8.98 (0.60) | 5.16 (0.65) | ||
| BART | 28.0% (5.6%) | 8.95 (1.20) | 9.15 (0.59) | 3.26 (0.43) | ||
| Lasso | 27.0% (3.9%) | 10.16 (1.04) | 9.10 (0.59) | 1.14 (0.08) | ||
| Boosting | 23.7% (3.6%) | 9.23 (1.10) | 9.05 (0.59) | 3.13 (0.38) | ||
| RLT-naive | 21.7% (4.0%) | 6.00 (1.86) | 8.71 (0.71) | 5.24 (0.64) | ||
|
| ||||||
| RLT
|
||||||
| Muting | Linear combination | |||||
|
|
||||||
| No | 1 | 17.8% (4.0%) | 4.93 (1.20) | 6.96 (0.98) | 4.89 (0.62) | |
| 2 | 20.3% (4.0%) | 6.09 (1.40) | 7.09 (0.85) | 3.35 (0.52) | ||
| 5 | 22.6% (3.9%) | 6.88 (1.53) | 7.25 (0.83) | 3.27 (0.52) | ||
| Moderate | 1 | 14.9% (3.9%) | 3.88 (1.11) | 6.43 (1.08) | 3.69 (0.47) | |
| 2 | 16.5% (4.3%) | 4.53 (1.32) | 6.47 (0.98) | 2.48 (0.36) | ||
| 5 | 18.6% (4.0%) | 5.26 (1.51) | 6.54 (0.96) | 2.40 (0.36) | ||
| Aggressive | 1 | 12.8% (3.8%) | 3.39 (1.04) | 6.13 (1.09) | 3.35 (0.44) | |
| 2 | 13.5% (4.1%) | 3.45 (1.16) | 6.14 (1.06) | 2.01 (0.25) | ||
| 5 | 14.8% (4.0%) | 4.09 (1.41) | 6.11 (1.05) | 1.89 (0.24) | ||
For each scenario, the best two methods within each panel are bolded. The overall best method is underlined.