Table 3.

Performance comparison of the proposed CNN-LSTM model with other classifiers.

Classifier	Datasets	Accuracy (95%CI)	F1 (95%CI)	AUC (95%CI)
CNN-LSTM	Training	0.970 (0.960–0.979)	0.970 (0.961–0.978)	0.975 (0.962–0.988)
	Validating	0.970 (0.962–0.977)	0.970 (0.960–0.976)	0.970 (0.961–0.978)
	Testing	0.970 (0.961–0.977)	0.970 (0.960–0.977)	0.975 (0.962–0.988)
LR	Training	0.655 (0.624–0.681)	0.655 (0.623–0.679)	0.656 (0.625–0.678)
	Validating	0.652 (0.621–0.674)	0.653 (0.622–0.675)	0.650 (0.623–0.671)
	Testing	0.621 (0.590–0.648)	0.621 (0.591–0.647)	0.621 (0.593–0.646)
DT	Training	0.600 (0.572–0.628)	0.697 (0.669–0.725)	0.668 (0.640–0.692)
	Validating	0.600 (0.571–0.627)	0.650 (0.595–0.695)	0.600 (0.570–0.628)
	Testing	0.591 (0.562–0.617)	0.611 (0.582–0.638)	0.591 (0.563–0.616)
KNN	Training	0.647 (0.609–0.676)	0.648 (0.602–0.686)	0.708 (0.673–0.744)
	Validating	0.600 (0.570–0.629)	0.600 (0.571–0.628)	0.600 (0.572–0.627)
	Testing	0.644 (0.618–0.669)	0.644 (0.619–0.668)	0.644 (0.617–0.667)
SVM	Training	0.880 (0.856–0.902)	0.880 (0.855–0.903)	0.880 (0.857–0.901)
	Validating	0.855 (0.828–0.877)	0.857 (0.834–0.892)	0.855 (0.829–0.878)
	Testing	0.745 (0.712–0.774)	0.745 (0.713–0.773)	0.771 (0.725–0.826)
RF	Training	0.897 (0.878–0.914)	0.893 (0.876–0.915)	0.867 (0.842–0.891)
	Validating	0.700 (0.671–0.726)	0.700 (0.672–0.725)	0.700 (0.673–0.728)
	Testing	0.788 (0.750–0.833)	0.788 (0.737–0.872)	0.788 (0.728–0.844)
AdaBoost	Training	0.636 (0.600–0.668)	0.647 (0.604–0.682)	0.670 (0.626–0.707)
	Validating	0.662 (0.633–0.688)	0.662 (0.647–0.688)	0.664 (0.625–0.691)
	Testing	0.658 (0.630–0.682)	0.665 (0.632–0.691)	0.655 (0.621–0.683)
GBDT	Training	0.798 (0.774–0.822)	0.797 (0.773–0.824)	0.884 (0.861–0.901)
	Validating	0.731 (0.704–0.756)	0.700 (0.671–0.726)	0.730 (0.702–0.755)
	Testing	0.611 (0.582–0.638)	0.608 (0.579–0.635)	0.682 (0.651–0.710)