Table 3.
Accuracy performance indicators for European universities’ co-publication prediction in Social Sciences and Humanities domain.
| Model | All | Exogenous | Endogenous |
|---|---|---|---|
| Best algorithm | GBM | RF | GBM |
| ROC | 0.9153 | 0.8945 | 0.9111 |
| Accuracy | 0.8331 | 0.8102 | 0.8288 |
| 95% CI | (0.8291, 0.837) | (0.806, 0.8144) | (0.8248, 0.8328) |
| No information rate | 0.8962 | 0.8947 | 0.8973 |
| P-value [Acc > NIR] | 1.0000 | 1.0000 | 1.0000 |
| Sensitivity | 0.8324 | 0.8091 | 0.8278 |
| Specificity | 0.8390 | 0.8196 | 0.8378 |
| Pos pred value | 0.9781 | 0.9744 | 0.9781 |
| Neg pred value | 0.3671 | 0.3356 | 0.3575 |
| Prevalence | 0.8962 | 0.8947 | 0.8973 |
| Detection rate | 0.7460 | 0.7240 | 0.7428 |
| Detection prevalence | 0.7627 | 0.7429 | 0.7595 |
| Balanced accuracy | 0.8357 | 0.8144 | 0.8328 |
Three models considered: with all, with only exogenous, and with only endogenous features.