Table 3.

Performance of algorithms with missing data handled by k-nearest neighbor imputation.

Algorithm	KNNa imputation
	AUCb, mean (SD)	P value	APc, mean (SD)	P value	Sensitivity, mean (SD)	P value	Specificity, mean (SD)	P value
LRd	0.81 (0.04)	.02	0.87 (0.03)	.10	0.46 (0.10)	<.001	0.91 (0.02)	>.99
LR+CSLe	0.81 (0.04)	.005	0.87 (0.02)	.03	0.65 (0.09)	.001	0.81 (0.03)	.90
DTf	0.67 (0.05)	<.001	0.77 (0.04)	<.001	0.48 (0.09)	<.001	0.86 (0.03)	>.99
DT+CSL	0.66 (0.07)	<.001	0.76 (0.05)	<.001	0.45 (0.13)	<.001	0.86 (0.02)	>.99
RFg	0.81 (0.04)	.004	0.87 (0.03)	.13	0.03 (0.04)	<.001	1.00 (0.00)	>.99
RF+CSL	0.81 (0.04)	.004	0.87 (0.03)	.02	0.68 (0.08)	.04	0.78 (0.02)	.10
SVMh (RBFi)	0.77 (0.06)	<.001	0.84 (0.03)	<.001	0.08 (0.04)	<.001	0.99 (0.01)	>.99
SVM (RBF)+CSL	0.80 (0.05)	<.001	0.86 (0.03)	<.001	0.75 (0.09)	.90	0.73 (0.02)	<.001
SVM (polyj)	0.75 (0.04)	<.001	0.83 (0.03)	<.001	0.50 (0.10)	<.001	0.86 (0.02)	>.99
SVM (poly)+CSL	0.81 (0.05)	<.001	0.86 (0.03)	<.001	0.74 (0.09)	.83	0.73 (0.02)	<.001
SVM (linear)	0.80 (0.04)	.001	0.86 (0.03)	.005	0.48 (0.11)	<.001	0.89 (0.02)	>.99
SVM (linear)+CSL	0.77 (0.04)	<.001	0.85 (0.02)	<.001	0.58 (0.11)	<.001	0.80 (0.02)	.75
SNNk	0.81 (0.06)	<.001	0.87 (0.04)	.006	0.33 (0.10)	<.001	0.95 (0.01)	>.99
SNN+CSL	0.80 (0.06)	<.001	0.86 (0.03)	<.001	0.65 (0.11)	.01	0.81 (0.03)	.90
DNNl	0.83 (0.05)	.04	0.88 (0.03)	.08	0.35 (0.09)	<.001	0.96 (0.01)	>.99
DNN+CSL	0.84 (0.04)	N/Am	0.88 (0.03)	N/A	0.72 (0.10)	N/A	0.79 (0.04)	N/A

^aKNN: k-nearest neighbor.

^bAUC: area under the receiver operating characteristic curve.

^cAP: average precision.

^dLR: logistic regression.

^eCSL: cost-sensitive learning.

^fDT: decision tree.

^gRF: random forest.

^hSVM: support vector machine.

ⁱRBF: radial basis function kernel.

^jpoly: polynomial kernel.

^kSNN: single neural network.

^lDNN: dual neural network.

^mN/A: not applicable.