Table 3. Performance of artificial neural networks (ANN) and bootstrap aggregation of complex decision trees (BACDT) classifiers for SpO₂ prediction at 5 min following a ventilator setting change, on test datasets (see Table 2).

Avg/total: average accuracy of total classification values. In italics is the performance of the best predictive model obtained among the eight tested.

Balanced datasets	5minSpO2 class	ANN			BACDT
		Precision	Recall	F-score	Precision	Recall	F-score
Dataset 1	1	0.12	0.70	0.21	0.80	0.76	0.78
	2	0.16	0.43	0.23	0.61	0.56	0.59
	3	0.96	0.67	0.79	0.97	0.98	0.97
	Avg/total	0.88	0.65	0.73	0.94	0.94	0.94
Dataset 2	1	0.09	0.72	0.16	0.77	0.72	0.74
	2	0.09	0.47	0.16	0.57	0.53	0.55
	3	0.98	0.70	0.81	0.98	0.99	0.98
	Avg/total	0.93	0.69	0.78	0.96	0.97	0.97
Dataset 3	1	0.16	0.68	0.25	0.80	0.76	0.78
	2	0.26	0.42	0.33	0.67	0.62	0.65
	3	0.92	0.60	0.72	0.95	0.96	0.96
	Avg/total	0.80	0.58	0.65	0.91	0.91	0.91
Dataset 4	1	0.09	0.69	0.16	0.80	0.74	0.77
	2	0.12	0.47	0.19	0.58	0.54	0.56
	3	0.97	0.68	0.80	0.98	0.98	0.98
	Avg/total	0.92	0.67	0.76	0.96	0.96	0.96