Table 5.

Average F1 scores on different threshold of uncertainty score T values. The values in brackets represent the standard deviation (Std) in 5-fold cross validation. The proposed SASMOTE algorithm performs better on the average F1 score with respect to the value of T ranging from 25% to 60% while the SASMOTE without visible neighbors performs better with respect to the value of T = 75% and T = 90%

Model	$\mathit{T}\mathbf{=}\mathbf{25}\mathbf{\%}$ (Std)	$\mathit{T}\mathbf{=}\mathbf{50}\mathbf{\%}$ (Std)	$\mathit{T}\mathbf{=}\mathbf{60}\mathbf{\%}$ (Std)	$\mathit{T}\mathbf{=}\mathbf{75}\mathbf{\%}$ (Std)	$\mathit{T}\mathbf{=}\mathbf{90}\mathbf{\%}$ (Std)
SASMOTE	50.45 (9.12)	51.16 (6.32)	49.63 (7.22)	48.44 (7.04)	47.06 (6.65)
SASMOTE w/o visible	49.26 (8.76)	50.04 (8.22)	47.39 (7.48)	49.72 (9.67)	49.51 (7.31)