TABLE 13:
CHEER’s performance on MIMIC-III, PTBDB and EEG with (left-column) K features with highest mutual information (MI) to the class label as features of the poor dataset; and (right-column) K features with lowest mutual information to the class label as features of the poor dataset. K is set to 2 for MIMIC-III, 5 for PTBDB and 7 for EEG.
| Dataset | Highest MI Features | Lowest MI Features |
|---|---|---|
| MIMIC-III | 0.672 ± 0.044 | 0.657 ± 0.012 |
| PTBDB | 0.646 ± 0.133 | 0.639 ± 0.115 |
| EEG | 0.815 ± 0.064 | 0.807 ± 0.042 |