Safe Reinforcement Learning on T2D cohort without Pump (Aggregated across cohorts)
86.71TIR (%)CRPO
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CRPOShielding=Predictive Shield, Pump=No Pump2026.01 | 86.71 | 3.47 | 20.35 | 0.11 | 0 | |
| CUPShielding=Predictive Shield, Pump=No Pump2026.01 | 85 | 4.35 | 25.18 | 8.93 | 0.74 | |
| PPOLagShielding=Predictive Shield, Pump=No Pump2026.01 | 81.75 | 4.94 | 33.69 | 9.36 | 1.79 | |
| RCPOShielding=Predictive Shield, Pump=No Pump2026.01 | 81.1 | 4.98 | 30.33 | 9.07 | 2.09 | |
| CPOShielding=Predictive Shield, Pump=No Pump2026.01 | 80.17 | 5.31 | 26.6 | 9.69 | 0 | |
| FOCOPSShielding=Predictive Shield, Pump=No Pump2026.01 | 80.02 | 5.03 | 32.63 | 13.96 | 1.35 | |
| TRPOLagShielding=Predictive Shield, Pump=No Pump2026.01 | 74.06 | 6.42 | 36.17 | 13.95 | 3.59 | |
| PCPOShielding=Predictive Shield, Pump=No Pump2026.01 | 45.35 | 17.61 | 50.93 | 6.22 | 29.42 |