Table 1 Reward comparison among the proposed DQN method, one-size-fit-all, random forest, and experts’ treatment.
From: Learning the Dynamic Treatment Regimes from Medical Registry Data through Deep Q-network
Treatments | Method | Reward | 95% Confidence Interval |
|---|---|---|---|
AGVHD | DQN | 0.717 | (0.683, 0.729) |
one-size-fit-all | 0.693 | (0.659, 0.705) | |
Random forest | 0.677 | (0.666, 0.703) | |
experts’ treatment | 0.673 | (0.663, 0.694) | |
CGVHD | DQN | 0.706 | (0.678, 0.722) |
one-size-fit-all | 0.684 | (0.671, 0.712) | |
Random forest | 0.672 | (0.663, 0.713) | |
experts’ treatment | 0.671 | (0.661, 0.697) |
