Table 7.

The details of implementing bias mitigation for the neural network model using stochastic gradient descent.

Sociodemographic conditions			Optimal threshold		Recall		Specificity		Accuracy		Differencea
Sex			18.40								1.08 b
	Male			85.71 c		58.99		59.12
	Female			85.19		58.43		58.63
Marital status			24.10								3.69
	Never been married			76.47		87.13		87.07
	Other groups of marital status			75.00		89.35		89.25
Working condition			24.50								15.22
	Working >35 hours			66.67		95.01		94.91
	Other groups of working condition			71.88		85.00		84.90
Race			18.40								26.60
	White			89.66		58.64		58.85
	Black			66.67		62.25		62.27
Income			17.30								8.89
	An income of <US $20,000			90.00		44.83		45.28
	Other groups of income			87.10		50.82		51.02

^aDifference between recall and specificity after bias mitigation.

^bItalicized values indicate an improvement compared to the initial values (50% threshold).