Table 5.
Summary of the ELM models with the best performance measurements precision and recall trained on triplets of the four datasets.
| Dataset | ALL2 | CNS | Adeno | DLBCL |
|---|---|---|---|---|
| Precision(Triplets) | 16535 | 732 | 17169 | 14625 |
| MinPrecision(ELM) | 1.000 | 1.000 | 1.000 | 1.000 |
| MaxPrecision(ELM) | 1.000 | 1.000 | 1.000 | 1.000 |
| (Min, Max) ranks of best-precision 10 models | 18.667 | 10.667 | 1.333 | 11.667 |
| 20.333 | 15.333 | 24.000 | 18.667 | |
| Recall(Triplets) | 4906 | 1369 | 17613 | 198 |
| MinRecall(ELM) | 1.000 | 1.000 | 1.000 | 1.000 |
| MaxRecall(ELM) | 1.000 | 1.000 | 1.000 | 1.000 |
| (Min, Max) ranks of best-Recall 10 models | 19.333 | 2.000 | 25.000 | 19.667 |
| 24.667 | 4.000 | 28.333 | 26.333 |
The numbers of models with precisions and recalls larger than CutOff were collected for the four datasets ALL2/CNS/Adeno/DLBCL in the row “Precision(Triplets)” and “Recall(Triplets)”, respectively. CutOff is 0.800 for the two difficult datasets ALL2 and CNS, and 0.900 for the two easy datasets Adeno and DLBCL. The minimum and maximum precisions of the best ten ELM models with the precisions larger than the cutoff were listed in the rows “MinPrecision(ELM)” and “MaxPrecision(ELM)”. And the next row gave the averaged rankings of the 10 triplets with the best precisions. The last three rows were defined similarly for the performance measurements Recall.