Table 3.

Comparisons of different experimental configurations.

Methods	Batch size	Learning rate	Epoch	Optimizer	Activation function	Dropout	Test loss	Test accuracy
Proposed model	8	0.0002	50	Nadam	ReLu	0.1	0.4683	95.54%
	16	0.0001	50	Adam	H-swish	0.5	0.3297	94.73%
	32	0.0002	100	Nadam	ReLu	0.2	0.2971	95.89%
	8	0.0001	80	Adam	ReLu	0.3	0.4508	93.85%
	16	0.0002	50	Adam	H-swish	0.2	0.1929	99.21%
Experimental settings	16	0.0002	50	Adam	H-swish	0.2	–	–