Skip to main content

View full-text article in PMC

. 2020 Aug 31;8(8):e19870. doi: 10.2196/19870

Table 4.

Effect of fully connected layers on the proposed dual neural network plus cost-sensitive learning model.

Imputation and algorithm		AUC^a, mean (SD)	AP^b, mean (SD)	Sensitivity, mean (SD)	Specificity, mean (SD)
Mean
	DNN^c+CSL^d	0.84 (0.04)	0.88 (0.03)	0.73 (0.09)	0.80 (0.03)
	DNN+CSL with one FCL^e	0.83 (0.04)	0.88 (0.03)	0.73 (0.09)	0.79 (0.07)
	DNN+CSL with two FCLs	0.83 (0.05)	0.88 (0.03)	0.77 (0.11)	0.75 (0.04)
KNN^f
	DNN+CSL	0.84 (0.04)	0.88 (0.03)	0.72 (0.10)	0.79 (0.04)
	DNN+CSL with one FCL	0.83 (0.04)	0.88 (0.03)	0.71 (0.10)	0.77 (0.09)
	DNN+CSL with two FCLs	0.82 (0.05)	0.87 (0.03)	0.77 (0.12)	0.74 (0.03)

^aAUC: area under the receiver operating characteristic curve.

^bAP: average precision.

^cDNN: dual neural network.

^dCSL: cost-sensitive learning.

^eFCL: fully connected layer.

^fKNN: k-nearest neighbor.