Table 7.

Summary of the hyperparameters of the top 10 best hybrid models.

Hybrid Model	Input Layer	LSTM Units	LSTM Dropout	FC Layer 1	FC Layer 2	Learning Rate	Batch Size	Total Params
Xception-LSTM	$20\times 2048$	512	0.4	1024	1024	$6.55\times {10}^{-4}$	8	6,822,915
NASNetLarge-LSTM	$20\times 4032$	1024	0.5	64	256	$6.62\times {10}^{-4}$	12	20,796,483
DenseNet121-LSTM	$20\times 1024$	32	0.1	1024	1024	$9.08\times {10}^{-4}$	20	1,221,763
POCOVID-Net-3-LSTM	$5\times 512$	256	0.2	128	128	$8.18\times {10}^{-4}$	32	837,251
DenseNet201-LSTM	$5\times 1920$	1024	0.2	512	128	$2.54\times {10}^{-4}$	4	12,653,571
NASNetMobile-LSTM	$15\times 1056$	64	0.5	64	1024	$6.60\times {10}^{-4}$	12	360,771
POCOVID-Net-1-LSTM	$20\times 512$	256	0.2	256	512	$5.21\times {10}^{-4}$	32	986,371
POCOVID-Net-4-LSTM	$10\times 512$	512	0.1	128	512	$1.63\times {10}^{-3}$	24	2,232,451
ResNet152V2-LSTM	$10\times 2048$	1024	0	1024	128	$7.76\times {10}^{-6}$	4	13,768,195
ResNet152V2-LSTM	$20\times 2048$	512	0.1	128	256	$1.18\times {10}^{-5}$	8	5,344,387