Table 3.

Classification performance of the proposed approach compared with the relevant CNN-based methods.

Methods	Input Length	Network	Validation	F1N	F1A	F1O	F1P	F1_NAO	F_NAOP	Visual Interpretation
[56]	30 s	ResNet (34 layers)	5-fold CV	90.2	65.7	69.8	64.0	75.2	72.4	None
[57]	N/A	2D CNN with LSTM layer	5-fold CV	88.8	76.4	72.6	64.5	79.2	75.58	None
[54]	9, 15 s	DenseNet	5-fold CV	91	80	76	N/A	82	N/A	None
[55]	9–61 s	16-layer 1D residual CRNN	5-fold CV	91.9	85.8	81.6	N/A	86.4	N/A	None
[52]	30 s	1D CNN	5-fold CV	N/A	N/A	N/A	N/A	82.2	78.2	None
[53]	60.5 s	Modified ResNet	8:1:1 split	N/A	N/A	N/A	N/A	79.59	N/A	Included
[58]	9–61 s	Dense18+ for spectrogram	10-fold CV	89.29	79.18	72.25	52.50	80.24	73.31	Included
Proposed	9–61 s	Proposed BIT-CNN	5-fold CV	89.73	81.06	74.45	62.22	81.75	76.87	Included