Table 5.

CDR machine learning model performance

Naive Bayes	0	0.5	1+	0	0.5	0.5	1+	0	1+
Accuracy	80.1%			81.8%		94.0%		98.3%
CI	75.36 – 84.84%			77.2 – 86.4%		91.2 – 96.8%		96.8 – 99.8%
G-mean	81.5%	74.2%	74.2%	77.1%	77.1%	84.9%	84.9%	93.8%	93.8%
Sensitivity	92.2%	61.3%	76.0%	92.2%	64.5%	100.0%	72.0%	100%	88.0%
Specificity	72.0%	89.9%	98.8%	64.5%	92.2%	72.0%	100%	88.0%	100%
Decision Tree	0	0.5	1+	0	0.5	0.5	1+	0	1+
Accuracy	80.5%			81.0%		94.1%		97.7%
CI	75.79 – 85.2%			76.3 – 85.7%		91.3–96.9%		95.9– 99.5%
G-Mean	79.9%	73.4%	89.1%	75.8%	75.8%	94.1%	94.1%	95.6%	95.6%
Sensitivity	94.2%	58.1%	80.0%	92.2%	62.4%	94.6%	92.0%	100%	84.0%
Specificity	67.8%	92.7%	99.2%	62.4%	92.2%	92.0%	94.6%	84.0%	100%
Logistic Regression	0	0.5	1+	0	0.5	0.5	1+	0	1+
Accuracy	70.0%			71.9%		91.2%		98.8%
CI	64.6 – 75.5%			66.6 – 77.2%		87.8 – 94.6%		97.5 – 100.0%
G-Mean	67.1%	60.9%	81.9%	65.6%	65.6%	81.2%	81.2%	96.5%	96.5%
Sensitivity	85.5%	44.2%	68.8%	84.1%	51.2%	95.8%	68.8%	99.3%	93.8%
Specificity	52.7%	83.9%	97.6%	51.2%	84.1%	68.8%	95.8%	93.8%	99.3%

Note: CI = confidence interval; G-mean = geometric mean.