Table 2.
Relative feature importance.
| Regulariser (λ) | OpA | Opα min | VS | Gradα | FD | AUC (95% CI) |
|---|---|---|---|---|---|---|
| Lasso (λmin) | 487 | 360 | 356 | – | – | 0.67 (0.66 to 0.68) |
| Lasso (λ1SE) | 443 | 174 | 212 | – | – | |
| Elastic-net (λmin) | 451 | 371 | 382 | – | – | 0.67 (0.66 to 0.68) |
| Elastic-net (λ1SE) | 451 | 118 | 235 | – | – | |
| Lasso (λmin) | 441 | 402 | 451 | 488 | – | 0.73 (0.73 to 0.74) |
| Lasso (λ1SE) | 315 | 190 | 272 | 450 | – | |
| Elastic-net (λmin) | 463 | 438 | 455 | 491 | – | 0.73 (0.73 to 0.74) |
| Elastic-net (λ1SE) | 362 | 243 | 283 | 451 | – | |
| Lasso (λmin) | 385 | 377 | 445 | 441 | 487 | 0.76 (0.75 to 0.76) |
| Lasso (λ1SE) | 202 | 187 | 316 | 439 | 259 | |
| Elastic-net (λmin) | 407 | 411 | 466 | 455 | 494 | 0.76 (0.75 to 0.77) |
| Elastic-net (λ1SE) | 253 | 220 | 261 | 335 | 460 | |
| Orlando et al. 13 (Lasso) | – | – | – | – | √ | 0.67 (0.52 to 0.80) |
Relative feature importance calculated by the number of times a feature reached a non-zero coefficient per bootstrap, average bootstrap performances and performances reported in 13 using fractal dimensions to detect PDR. Bold indicates the most important feature.
CI confidence intervals, AUC area under curve.