Table 4. External Validation of Imbalanced Methods.
| combination
of hybrid and cluster under-sampling |
hybrid under-sampling |
multiple under-sampling |
||||
|---|---|---|---|---|---|---|
| assay AID | BA | GM | BA | GM | BA | GM |
| 651632 | 0.68 (AD: 100%) | 0.62 (AD: 100%) | 0.68 (AD: 100%) | 0.65 (AD: 100%) | 0.68 (AD: 100%) | 0.66 (AD: 100%) |
| 651632 (combined AD) | 0.68 (AD: 95.2%) | 0.63 (AD: 95.2%) | 0.69 (AD: 79.3%) | 0.66 (AD: 79.3%) | 0.69 (AD: 92.2%) | 0.67 (AD: 92.2%) |
AD: applicability domain. BA: balanced accuracy. GM: geometric mean (G-mean). Lower row: predictions limited to those that fall in the combined applicability domains of the hybrid and cluster combination methods.