Table 3. Comparison of performance estimates based on random cross-validation and cluster-cross-validation. The table provides the performances in terms of AUC averaged over the different targets for both random cross-validation and cluster-cross validation. The performance estimates obtained from random cross-validation are on average 0.02 higher, i.e., more optimistic, than the ones from cluster-cross-validation.
| Cluster-cross-validation | Fold 1: | 0.722 ± 0.138 | Fold 2: | 0.729 ± 0.148 | Fold 3: | 0.743 ± 0.147 |
| Random cross-validation | Fold I: | 0.747 ± 0.145 | Fold II: | 0.760 ± 0.136 | Fold III: | 0.754 ± 0.142 |