Table 2.

Performance of different methods on models in Perspectives #1 and #2

Metrics	Naïve Bayes	Bayes net	Logistic regression	Random forest	LR&RF_VotingEnsemble
Models built in Perspective #1—data collection window (past, AKI-onset - 1 day)
AUC	0.687 (0.686–0.687)	0.687 (0.687–0.687)	0.726 (0.725–0.726)	0.709 (0.708–0.710)	0.744 (0.743–0.744)
F-measure	0.261 (0.260–0.262)	0.262 (0.261–0.262)	0.317 (0.316–0.318)	0.317 (0.316–0.318)	0.330 (0.329–0.331)
Sensitivity (recall)	47.6% (42.5–52.7%)	47.5% (46.7–48.3%)	40.6% (39.8–41.4%)	40.7% (39.8–41.5%)	40.3% (39.4–41.1%)
Specificity	77.4% (76.8–77.9%)	77.6% (77.0–78.1%)	87.9% (87.5–88.4%)	87.9% (87.4–88.4%)	89.2% (88.8–89.6%)
Precision	18.0% (17.8–18.1%)	18.1% (17.9–18.2%)	26.1% (25.9–26.4%)	26.0% (25.6–26.4%)	28.0% (27.7–28.3%)
Models built in Perspective #2–data collection window (past, admission)
AUC	0.676 (0.676–0.676)	0.677 (0.677–0.677)	0.719 (0.718–0.720)	0.714 (0.713–0.715)	0.734 (0.734–0.735)
F-measure	0.253 (0.252–0.253)	0.253 (0.252–0.254)	0.308 (0.308–0.309)	0.294 (0.293–0.295)	0.318 (0.317–0.319
Sensitivity (recall)	45.4% (44.4–46.3%)	45.7% (44.8–46.6%)	40.3% (39.3–41.3%)	40.4% (39.7–41.1%)	40.6% (39.9–41.2%)
Specificity	77.4% (76.8–77.9%)	77.6% (77.0–78.1%)	87.9% (87.5–88.4%)	87.9% (87.4–88.4%)	89.2% (88.8–89.6%)
Precision	17.5% (17.4–17.6%)	17.5% (17.4–17.7%)	25.0% (24.6–25.3%)	23.1% (22.8–23.4%)	26.2% (25.8–26.5%)

Abbreviations: AKI: acute kidney injury; AUC: area under the curve.