A Long Short-Term Memory Ensemble Approach for Improving the Outcome Prediction in Intensive Care Unit

Table 2

Evaluations of all mortality prediction systems (mean ± std).

	SAPS II	SOFA	APACHE II	RF	LSTM	eLSTM	ANOVA test
AUROC	0.7788 ± 0.0166	0.7354 ± 0.0184	0.7467 ± 0.0173	0.8282 ± 0.0151	0.8382 ± 0.0158	0.8451 ± 0.0136	F = 926.328,
AUPRC	0.3800 ± 0.0334	0.3381 ± 0.0307	0.3515 ± 0.0306	0.4197 ± 0.0393	0.4751 ± 0.0351	0.4862 ± 0.0345	F = 426.683,
Sensitivity/recall	0.6922 ± 0.0267	0.5418 ± 0.0394	0.6478 ± 0.0303	0.7197 ± 0.0395	0.7384 ± 0.0401	0.7758 ± 0.0321	F = 438.869,
Specificity	0.7404 ± 0.0102	0.7958 ± 0.0101	0.7256 ± 0.0119	0.7807 ± 0.0218	0.7746 ± 0.0182	0.7503 ± 0.0136	F = 229.707,
Accuracy	0.7347 ± 0.0096	0.7658 ± 0.0106	0.7164 ± 0.0113	0.7734 ± 0.0174	0.7703 ± 0.0148	0.7533 ± 0.0112	F = 234.492,
Precision	0.2633 ± 0.0145	0.2622 ± 0.0179	0.2404 ± 0.0149	0.3063 ± 0.0211	0.3056 ± 0.0208	0.2941 ± 0.0158	F = 271.132,
F1	0.3813 ± 0.0180	0.3532 ± 0.0227	0.3505 ± 0.0187	0.4290 ± 0.0216	0.4317 ± 0.0230	0.4262 ± 0.0181	F = 363.817,

The difference with the eLSTM model is significant at the 0.05 level. Bold indicates the highest mean value.