Table 4.

Performance metrics for the final models. The means with 95% confidence intervals in parentheses are represented. RF was the top performing traditional machine learning model. LSTM: long short-term memory; AP: average precision; AUROC: area under the receiver operating characteristic curve.

Metric	Hourly LSTM	Daily LSTM	Neural Network	Random Forest
AP	0.289 (0.287, 0.292)	0.570 (0.567, 0.573)	0.526 (0.523, 0.529)	0.552 (0.549, 0.555)
Sensitivity	0.962 (0.960, 0.963)	0.963 (0.961, 0.964)	0.768 (0.765, 0.771)	0.334 (0.331, 0.337)
Specificity	0.431 (0.430, 0.431)	0.238 (0.237, 0.238)	0.893 (0.892, 0.893)	0.995 (0.995, 0.996)
AUROC	0.854 (0.852, 0.855)	0.905 (0.904, 0.906)	0.901 (0.899, 0.902)	0.899 (0.898, 0.900)