Table 2.

Classification metrics on the test dataset using the different architecture of deep transfer learning models and also proposed ensemble method. For each model, average (± std.) performance measure is reported over the best 5 trained model checkpoints

Model	Precision	Recall	F1-score	Accuracy	AUC
EfficientNetB0	0.847(± 0.03)	0.822(± 0.11)	0.815(± 0.05)	0.82(± 0.02)	0.907(± 0.02)
EfficientNetB1	0.727(± 0.06)	0.718(± 0.09)	0.712(± 0.03)	0.71(± 0.02)	0.809(± 0.02)
EfficientNetB2	0.768(± 0.03)	0.768(± 0.12)	0.768(± 0.05)	0.77(± 0.03)	0.859(± 0.03)
EfficientNetB3	0.769(± 0.03)	0.765(± 0.07)	0.763(± 0.03)	0.76(± 0.03)	0.851(± 0.01)
EfficientNetB4	0.791(± 0.02)	0.789(± 0.05)	0.788(± 0.01)	0.79(± 0.01)	0.877(± 0.01)
EfficientNetB5	0.817(± 0.03)	0.817(± 0.11)	0.817(± 0.05)	0.82(± 0.03)	0.886(± 0.01)
Inception_resnet_v2	0.773(± 0.03)	0.774(± 0.12)	0.773(± 0.05)	0.77(± 0.02)	0.856(± 0.01)
InceptionV3	0.825(± 0.03)	0.814(± 0.07)	0.815(± 0.03)	0.82(± 0.02)	0.897(± 0.02)
NASNetLarge	0.772(± 0.06)	0.770(± 0.09)	0.768(± 0.03)	0.77(± 0.01)	0.836(± 0.03)
NASNetMobile	0.759(± 0.03)	0.757(± 0.12)	0.757(± 0.05)	0.76(± 0.04)	0.823(± 0.02)
ResNet50	0.807(± 0.03)	0.808(± 0.11)	0.807(± 0.05)	0.81(± 0.03)	0.875(± 0.01)
Xception	0.738(± 0.06)	0.739(± 0.09)	0.738(± 0.03)	0.74(± 0.04)	0.782(± 0.04)
DenseNet121	0.768(± 0.03)	0.768(± 0.03)	0.768(± 0.03)	0.77(± 0.02)	0.868(± 0.04)
SeResnet50	0.755(± 0.03)	0.745(± 0.07)	0.745(± 0.03)	0.75(± 0.02)	0.818(± 0.02)
ResNext50	0.810(± 0.03)	0.806(± 0.12)	0.806(± 0.05)	0.81(± 0.02)	0.843(± 0.02)
Proposed ensemble model	0.857(± 0.02)	0.854(± 0.05)	0.852(± 0.01)	0.852(± 0.01)	0.91(± 0.01)