Table 2.

Architecture of selected models tested in this study. The miniCNN, smallCNN, mediumCNN, largeCNN, and verylargeCNN models have one 1, 2, 3, 4, and 7 CNN layers, respectively. The number of kernels, and their sizes (within parentheses) are indicated under column ‘Conv kernels’. The pooling size is indicated under ‘Pooling size’. The three smallRNN models had one LSTM layer each. The sizes of LSTM nodes are indicated under column ‘LSTM size’. Pooling layer after the last CNN layer was based in ‘Global Max’, and pooling layer for other layers a local max pooling function of size 10 was used.

Model	Conv Layers	Conv kernels in each layer (length)	Pooling size	LSTM Layers	LSTM Size
16_miniCNN	1	16 (15)	Global Max	0	0
32_miniCNN	1	32 (15)	Global Max	0	0
64_miniCNN	1	64 (15)	Global Max	0	0
16_smallCNN	2	16 (15)	10, Global Max	0	0
32_smallCNN	2	32 (15)	10, Global Max	0	0
64_smallCNN	2	64 (15)	10, Global Max	0	0
16_mediumCNN	3	16 (15)	10, Global Max	0	0
32_mediumCNN	3	32 (15)	10, Global Max	0	0
64_mediumCNN	3	64 (15)	10, Global Max	0	0
16_largeCNN	4	16 (15)	10, Global Max	0	0
32_largeCNN	5	32 (15)	10, Global Max	0	0
64_largeCNN	5	64 (15)	10, Global Max	0	0
16_verylargeCNN	7	16 (15)	10, Global Max	0	0
32_verylargeCNN	7	32 (15)	10, Global Max	0	0
64_verylargeCNN	7	64 (15)	10, Global Max	0	0
16_smallRNN	0	0	N/A	1	15
5_smallRNN	0	0	N/A	1	5
30_smallRNN	0	0	N/A	1	30