Table 4 TCGA breast cancer data (N = 960, events = 196). The prediction performance of stacking and the other methods (mean (SD)). Results are based on 100 random splits of the original data to training set (N = 480) and test set (N = 480) (19 candidate pathways)
| Â | Deviance | AUC | Brier score | Misclassification |
|---|---|---|---|---|
Penalty and group penalty methods | ||||
Lasso | 479.216 (20.363) | 0.587 (0.037) | 0.160 (0.008) | 0.203 (0.014) |
MCP | 481.482 (23.224) | 0.575 (0.045) | 0.161 (0.009) | 0.203 (0.015) |
SCAD | 481.901 (26.133) | 0.586 (0.040) | 0.160 (0.008) | 0.203 (0.014) |
gsslasso | 479.552 (21.890) | 0.594 (0.031) | 0.160 (0.009) | 0.203 (0.013) |
grlasso | 481.971 (18.018) | 0.554 (0.035) | 0.161 (0.008) | 0.203 (0.013) |
grMCP | 487.903 (26.992) | 0.537 (0.033) | 0.163 (0.009) | 0.204 (0.015) |
grSCAD | 481.545 (18.039) | 0.559 (0.034) | 0.161 (0.008) | 0.202 (0.014) |
cMCP | 497.776 (49.826) | 0.577 (0.038) | 0.165 (0.013) | 0.210 (0.021) |
Model stacking methods | ||||
nsslasso (Lasso) | 481.597 (20.492) | 0.598 (0.027) | 0.161 (0.008) | 0.207 (0.013) |
nsslasso (network) | 485.152 (16.481) | 0.583 (0.010) | 0.163 (0.007) | 0.208 (0.011) |
nsslasso (KNN) | 488.225 (21.950) | 0.549 (0.035) | 0.163 (0.009) | 0.204 (0.013) |
nsslasso (NBa) | 489.239 (19.413) | 0.531 (0.022) | 0.163 (0.008) | 0.202 (0.014) |
nsslasso (RF) | 486.269 (20.569) | 0.551 (0.032) | 0.162 (0.008) | 0.203 (0.013) |
L-BFGS (Lasso) | 476.821 (18.592) | 0.602 (0.026) | 0.159 (0.008) | 0.202 (0.013) |
nLasso (Lasso) | 480.495 (18.962) | 0.583 (0.042) | 0.161 (0.008) | 0.205 (0.013) |