TABLE 3.
Effect of sample sizes n on DNN’s performance in terms of c-index (mean and SD from 200 replications) in the presence of high-dimensional predictors
| n | ||||||
|---|---|---|---|---|---|---|
| p | 200 | 500 | 1000 | 1500 | 2000 | |
| Scenario 4 | 100 | 65.4 (4.3) | 78.9 (2.0) | 84.5 (2.0) | 87.8 (1.7) | 89.0 (1.6) |
| 500 | 57.2 (3.2) | 62.3 (3.0) | 76.3 (1.8) | 79.9 (1.7) | 82.4 (1.1) | |
| Scenario 5 | 100 | 66.7 (4.7) | 82.8 (1.8) | 88.6 (0.8) | 89.6 (0.6) | 90.5 (0.5) |
| 500 | 58.2 (3.3) | 63.2 (3.4) | 80.4 (2.0) | 86.0 (2.4) | 87.8 (1.0) |
Note: Both scenarios (4 and 5) are from the sparse signal setting. The number of predictors is set at p = 100 or 500.
Abbreviation: DNN, deep neural network.