Table 2.
Model performance between no formal report and 2500 formal report based on five metrics (the highest value for each metric is highlighted in bold type): multi-instance learning methods outperformed baselines
| Method | Formal | ACC | PR | RE | FS | AUC |
|---|---|---|---|---|---|---|
| #Report | ||||||
| SVM(linear) | 0 | 0.7793 | 0.7309 | 0.6100 | 0.6644 | 0.7916 |
| 2500 | 0.7296 | 0.6241 | 0.6370 | 0.6294 | 0.7234 | |
| SVM(poly) | 0 | 0.6412 | 0.7231 | 0.3611 | 0.3069 | 0.5697 |
| 2500 | 0.5478 | 0.5311 | 0.5497 | 0.4443 | 0.6416 | |
| SVM(rbf) | 0 | 0.6507 | 0.6948 | 0.0572 | 0.1035 | 0.8069 |
| 2500 | 0.5897 | 0.4652 | 0.9344 | 0.6210 | 0.7754 | |
| LR | 0 | 0.7665 | 0.6765 | 0.6641 | 0.6700 | 0.7524 |
| 2500 | 0.7322 | 0.6209 | 0.6576 | 0.6384 | 0.7303 | |
| NN | 0 | 0.7924 | 0.7408 | 0.6273 | 0.6790 | 0.8196 |
| 2500 | 0.7411 | 0.6414 | 0.6396 | 0.6394 | 0.7366 | |
| miFV | 0 | 0.7818 | 0.7269 | 0.6352 | 0.6775 | 0.8348 |
| 2500 | 0.7856 | 0.7331 | 0.6403 | 0.6833 | 0.8361 | |
| miVLAD | 0 | 0.7691 | 0.7261 | 0.5832 | 0.6461 | 0.8390 |
| 2500 | 0.7863 | 0.7055 | 0.6999 | 0.7018 | 0.8201 | |
| MILR | 0 | 0.8034 | 0.7858 | 0.6231 | 0.6947 | 0.8676 |
| 2500 | 0.8054 | 0.7871 | 0.6291 | 0.6984 | 0.8902 |