Table 2.

DeepWrist’s performance on trivial cases (test set #1) and hard cases ( test set #2).

Dataset	Model	AUROC (95% CI)	AUPR (95% CI)	Sensitivity, recall, TPR (95% CI)	Specificity, selectivity, TNR (95% CI)	Precision PPV (95% CI)	$F_1$ score (95% CI)	BA (95% CI)
Test set #1	PA	0.98 (0.97–0.99)	0.99 (0.98–0.99)	0.97 (0.94–1.00)	0.88 (0.80–0.94)	0.93 (0.89–0.96)	0.95 (0.92–0.97)	0.93 (0.89–0.96)
	LAT	0.98 (0.97–0.99)	0.99 (0.98–0.99)	0.97 (0.94–1.00)	0.91 (0.84–0.96)	0.94 (0.91–0.97)	0.96 (0.93–0.98)	0.94 (0.90–0.97)
	Ensemble	0.99 (0.98–0.99)	0.99 (0.99–0.99)	0.97 (0.94–1.00)	0.87 (0.79–0.93)	0.92 (0.88–0.96)	0.95 (0.92–0.97)	0.92 (0.88–0.96)
Test set #2	PA	0.81 (0.69–0.91)	0.61 (0.44–0.80)	0.50 (0.30–0.70)	0.89 (0.82–0.95)	0.52 (0.33–0.73)	0.51 (0.31–0.68)	0.69 (0.58–0.80)
	LAT	0.83 (0.70–0.93)	0.57 (0.41–0.80)	0.50 (0.30–0.70)	0.94 (0.88–0.98)	0.66 (0.46–0.90)	0.57 (0.36–0.75)	0.72 (0.60–0.83)
	Ensemble	0.84 (0.72–0.93)	0.64 (0.46–0.83)	0.60 (0.40–0.80)	0.92 (0.87–0.97)	0.66 (0.48–0.87)	0.63 (0.44–0.80)	0.76 (0.65–0.87)

Here, AUROC is area under the receiver operating characteristic, AUPR is the area under precision recall curve, CI is confidence interval, 95% CI is shown in parentheses, TPR is true positive rate, TNR is true negative rate and PPV is positive predictive value and BA stands for balanced accuracy.