Table 1 Performance metrics for different models across tasks T₁ to T₆

Task	Model	Accuracy	Precision	Recall	f1-score	Wilcoxon test^a
T₁	Basic Tr.	0.77	0.74	0.77	0.73	4.64e−20
	Stacked Tr.	0.77	0.72	0.77	0.73	4.65e−20
	CNN Tr.	0.75	0.69	0.74	0.70	5.38e−21
	Dual Stream Tr.	0.86	0.87	0.86	0.85	–
T₂	Basic Tr.	0.72	0.66	0.72	0.67	6.61e−9
	Stacked Tr.	0.73	0.67	0.73	0.68	1.44e−8
	CNN Tr.	0.73	0.67	0.73	0.68	1.34e−8
	Dual Stream Tr.	0.77	0.79	0.77	0.73	–
T₃	Basic Tr.	0.78	0.72	0.78	0.73	6.48e−18
	Stacked Tr.	0.77	0.72	0.77	0.73	3.93e−14
	CNN Tr.	0.67	0.61	0.67	0.63	5.99e−24
	Dual Stream Tr.	0.78	0.79	0.78	0.77	–
T₄	Basic Tr.	0.77	0.72	0.77	0.72	2.32e−12
	Stacked Tr.	0.77	0.71	0.77	0.72	2.32e−12
	CNN Tr.	0.70	0.74	0.70	0.65	5.47e−20
	Dual Stream Tr.	0.77	0.77	0.77	0.74
T₅	Basic Tr.	0.75	0.70	0.75	0.70	5.87e−17
	Stacked Tr.	0.76	0.71	0.76	0.72	4.01e−16
	CNN Tr.	0.70	0.74	0.70	0.65	9.65e−22
	Dual Stream Tr.	0.80	0.81	0.80	0.78
T₆	Basic Tr.	0.75	0.69	0.75	0.70	8.73e−19
	Stacked Tr.	0.76	0.70	0.76	0.71	4.73e−18
	CNN Tr.	0.71	0.66	0.71	0.66	9.46e−22
	Dual Stream Tr.	0.78	0.80	0.78	0.76

^aThe p-values in the “Wilcoxon test” column indicate the outcomes of the Wilcoxon Signed-Rank test conducted pairwise comparing the Dual Stream Transformer and the other models (Basic Transformer, Stacked Transformer, and CNN Transformer). A p-value below <0.05 signifies a statistically significant disparity in performance, with the Dual Stream Transformer consistently surpassing the other models. Best performance metrics are highlighted in bold.

Quick links

Search