Table 3.

Average performance of algorithms across trials with full information. The table presents means and standard errors separately for the four algorithms with the training set (fitting) and test set (prediction). For each combination, the table reports the mean and standard error of the mean for sensitivity, specificity and balanced accuracy across trials. FFTi, UDT and LogReg were tested in 10,000 trials and FFTd in the first 1000 of these only.

		Training			Prediction
		Sensitivity	Specificity	Bal. Accuracy	Sensitivity	Specificity	Bal. Accuracy
FFTi	M	0.767	0.723	0.745	0.698	0.695	0.696
	SE	(0.0008)	(0.0008)	(0.0002)	(0.0011)	(0.0009)	(0.0003)
FFTd	M	0.815	0.756	0.786	0.698	0.71	0.704
	SE	(0.0025)	(0.0024)	(0.0007)	(0.0032)	(0.0028)	(0.0010)
UDT	M	0.896	0.784	0.840	0.625	0.694	0.660
	SE	(0.0005)	(0.0005)	(0.0002)	(0.0010)	(0.0007)	(0.0004)
LogReg	M	0.792	0.803	0.798	0.632	0.749	0.690
	SE	(0.0004)	(0.0003)	(0.0003)	(0.0007)	(0.0004)	(0.0003)

M—mean, SE—standard error of the mean, FFTi—fast-and-frugal tree construction using the ifan algorithm, FFTd—fast-and-frugal tree construction using the dfan algorithm, UDT—unconstrained decision tree based on the CART algorithm, LogReg—logistic regression.