Table 3.

Classification performance of our proposed ensemble-SDCNN^a model including the submodels as an ablation study.

Data sets and models		F1	APb	ARc	ACCd	AUCe
MCf
	SCNNg,h	0.765	0.775	0.757	0.769	0.817
	DCNNi,j	0.88	0.888	0.872	0.878	0.932
	ensemble-SDCNN	0.929	0.937	0.921	0.928	0.965
SZk
	SCNN	0.802	0.803	0.802	0.802	0.868
	DCNN	0.892	0.892	0.892	0.891	0.939
	ensemble-SDCNN	0.908	0.909	0.908	0.908	0.948
MC + SZ
	SCNN	0.79	0.793	0.788	0.789	0.841
	DCNN	0.891	0.892	0.89	0.89	0.943
	ensemble-SDCNN	0.9	0.902	0.898	0.899	0.95
MC train and SZ test
	SCNN	0.557	0.559	0.555	0.557	0.541
	DCNN	0.54	0.574	0.51	0.517	0.737
	ensemble-SDCNN	0.795	0.798	0.793	0.792	0.853
SZ train and MC test
	SCNN	0.625	0.624	0.626	0.616	0.601
	DCNN	0.7	0.702	0.698	0.71	0.754
	ensemble-SDCNN	0.811	0.808	0.813	0.797	0.873

^aSDCNN: shallow–deep CNN.

^bAP: average precision.

^cAR: average recall.

^dACC: accuracy.

^eAUC: area under the curve.

^fMC: Montgomery County.

^gAblation study performance by only considering SCNN for classification.

^hSCNN: shallow CNN.

ⁱAblation study performance by only considering DCNN for classification.

^jDCNN: deep CNN.

^kSZ: Shenzhen.