Table 2.

Machine learning prediction of benign and nonbenign cytology

	Accuracy, % (± SD)	False-negative rate, % (± SD)	False-positive rate, % (± SD)	Area under receiver operator curve, % (± SD)
Historical practice (clinical formulation)	65.67	0.00	82.14	58.93
Random forest	83.55 ± 1.58	12.50 ± 4.79	21.43 ± 9.22	83.04 ± 2.48
Rule set	77.57 ± 5.07	10.27 ± 6.78	14.10 ± 5.43	83.78 ± 4.46

Compares the results of historical practice to random forest classifier and the simplified rule set using 4 measures of performance.