Table 4.
Accuracy performance indicators for European universities’ co-publication prediction in Physical and Engineering Sciences.
| Model | All | Exogenous | Endogenous |
|---|---|---|---|
| Best algorithm | GBM | RF | GBM |
| ROC | 0.9416 | 0.9173 | 0.9370 |
| Accuracy | 0.8668 | 0.8406 | 0.8691 |
| 95% CI | (0.8632, 0.8703) | (0.8368, 0.8444) | (0.8655, 0.8726) |
| No information rate | 0.8012 | 0.7978 | 0.8007 |
| P-value [Acc > NIR] | 0.0000 | 0.0000 | 0.0000 |
| Sensitivity | 0.8695 | 0.8418 | 0.8745 |
| Specificity | 0.8562 | 0.8361 | 0.8474 |
| Pos pred value | 0.9606 | 0.9530 | 0.9584 |
| Neg pred value | 0.6194 | 0.5725 | 0.6270 |
| Prevalence | 0.8012 | 0.7978 | 0.8007 |
| Detection rate | 0.6966 | 0.6716 | 0.7002 |
| Detection prevalence | 0.7252 | 0.7047 | 0.7306 |
| Balanced accuracy | 0.8628 | 0.8389 | 0.8609 |
Three models considered: with all, with only exogenous, and with only endogenous features.