Table 4. Performance of the DL, RF and GLM models and backcasting of the LEMA 0–8, EMA 8–24, EMA 24–48, and EMA 48–72 datasets.

	AUC (95% CI)	Specificity	Sensitivity	Accuracy
DL-LEMA 0–8	0.982(0.973–0.991)	0.947	0.935	0.945
RF-LEMA 0–8	0.968(0.953–0.983)	0.938	0.921	0.936
GLM-LEMA 0–8	0.966(0.95–0.981)	0.939	0.907	0.934
DL-EMA 8–24	0.894(0.868–0.914)	0.840	0.805	0.815
RF-EMA 8–24	0.872(0.846–0.898)	0.852	0.781	0.838
GLM-EMA 8–24	0.874(0.848–0.90)	0.863	0.771	0.844
DL-EMA 8–24 Backcasting	0.898(0.875–0.921)	0.827	0.832	0.828
RF-EMA 8–24 Backcasting	0.888(0.862–0.913)	0.859	0.808	0.849
GLM-EMA 8–24 Backcasting	0.874(0.847–0.848)	0.856	0.771	0.838
DL-EMA 24–48	0.873(0.840–0.907)	0.821	0.767	0.811
RF-EMA 24–48	0.897(0.926–0.926)	0.845	0.797	0.836
GLM-EMA 24–48	0.873(0.838–0.908)	0.813	0.827	0.815
DL-EMA 24–48 Backcasting	0.892(0.867–0.928)	0.838	0.839	0.838
RF-EMA 24–48 Backcasting	0.896(0.865–0.928)	0.870	0.821	0.860
GLM-EMA 24–48 Backcasting	0.862(0.825–0.899)	0.827	0.785	0.819
DL-EMA 48–72	0.846(0.812–0.902)	0.797	0.757	0.792
RF-EMA 48–72	0.841(0.807–0.875)	0.814	0.733	0.804
GLM-EMA 48–72	0.839(0.803–0.875)	0.758	0.787	0.762
DL-EMA 48–72 Backcasting	0.871(0.841–0.902)	0.820	0.787	0.816
RF-EMA 48–72 Backcasting	0.850(0.814–0.887)	0.849	0.751	0.836
GLM-EMA 48–72 Backcasting	0.818(0.78–0.857)	0.764	0.727	0.759