Table 1.
Summary of the dataset and tasks used to evaluate Deep Explanation Ensembles alongside baseline model and training hyperparameters.
| Dataset name | Dataset descriptors | Baseline model hyperparameters | Baseline training hyperparameters | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Task | Num. samples | Num. features | Num. classes | Model architecture | Num. hidden layers | Num. epochs | Learning rate | Batch size | |
| Breast cancer Wisconsin | Binary classification | 569 | 10 | 2 | MLP | 2 | 14 | 0.001 | 64 |
| KAIMRC | Binary classification | 18,844 | 24 | 2 | MLP | 3 | 14 | 0.001 | 32 |
| KAIMRC | Regression | 18,844 | 24 | N/A | MLP | 3 | 14 | 0.001 | 32 |
| MIMIC-IV | Binary classification | 383,220 | N/A | 2 | flexible-ehr | 4 | 20 | 0.0005 | 128 |
| Codon usage (Kingdom) | Multi-class classification | 12,964 | 64 | 5 | MLP | 1 | 16 | 0.0001 | 32 |
| Codon usage (DNA) | Multi-class classification | 12,964 | 64 | 3 | MLP | 1 | 20 | 0.0001 | 32 |
Note that MIMI-IV is a time-series dataset and so each entry will have different numbers of features, and the KAIMRC (Regression) task has no target class as it is a regression problem.