. 2023 Sep 15;23(18):7924. doi: 10.3390/s23187924

Table 5.

Performance comparison on the OSASUD dataset considering per second and per-patient.

			Per Second
Model	Base Architecture	Dataset	AUC	Acc *	F1	Sens *	Spec *	Prec *
Mostafa et al. [33]	CNN	Private	-	89.40	-	74.40	93.90	-
Mostafa et al. [33]	CNN	UCD	-	66.79	-	85.37	60.94	-
Morillo et al. [22]	PNN	Private	0.889	85.22	-	92.4	95.9	-
Bernardini et al. [28]	LSTM	OSAUCD	0.704	0.676	0.399	0.656	0.680	-
proposed	Transformer	OSAUCD	0.908	0.821	0.769	0.808	0.829	0.733
			Per-patient (OSA = AHI ≥ 5)
Bernardini et al. [28]	LSTM	OSAUCD	-	0.633	0.776	0.826	0.0	-
Proposed (AVG *)	Transformer	OSAUCD	0.868	0.804	0.422	0.647	0.801	0.385
Proposed (MAX *)	Transformer	OSAUCD	0.990	0.968	0.826	0.996	0.992	0.776
Proposed (MIN *)			0.553	0.522	0.034	0.061	0.235	0.0176
Proposed (σ *)			0.119	0.121	0.254	0.295	0.162	0.249

* Abbreviations: Acc, accuracy; Sens, sensitivity; Spec, specificity; Prec, precision; AVG, Average; MAX, maximum; MIN, minimum; σ, Standard deviation.