Table 4. Performance of the four machine-learning models for PIH prediction.

		Naïve Bayes	Logistic regression	Random forest	ANN
All features	accuracy	55.74	62.01	76.28	70.7
	precision	80	70.25	77.99	74.51
	recall	13.33	60.79	81.28	74.5
	AUC	60.16	60.47	79.5	76.01
	95% CI	45.41–74.62	46.69–74.26	67.87–91.14	64.03–88
Feature set A (Remove redundant features)	accuracy	53.64	59.97	68.28	64.53
	precision	78.5	67.11	71.36	69.14
	recall	28.02	59.23	74.61	69.27
	AUC	67.23	66.78	71.85	70.72
	95% CI	53.19–81.27	52.74–80.81	58.58–85.1	56.98–84.47
Feature set B (Rank features by importance)	accuracy	70.2	79.16	78.8	70.61
	precision	71.54	79.95	79.5	72.92
	recall	79.71	85.01	84.58	77.64
	AUC	77.82	75.56	83.78	67.57
	95% CI	65.87–89.76	63.02–88.11	73.36–94.2	53.53–81.6
Feature set C (Recursive feature elimination)	accuracy	70.02	68.56	79.48	68.62
	precision	77.08	72.75	81.16	72.25
	recall	67.13	71.67	83.65	72.97
	AUC	77.25	73.42	84.23	72.3
	95% CI	65.13–89.38	60.57–86.27	73.63–94.84	59.5–85.09

ANN, artificial neural network; AUC, area under receiver operating characteristic curve. Table notes the precision and recall for the hypotension class.