Table 2.

Performance measures for the four classifiers

	Accuracy	Sensitivity	Specificity	PPV	NPV	F1	AUC
Training: melanoma (LOOCV estimate)
Classification trees	0.794	0.840	0.774	0.667	0.914	0.753	0.748
LASSO-LR	0.803	0.905	0.750	0.655	0.938	0.760	0.827
Random forest	0.754	0.762	0.750	0.615	0.857	0.681	0.756
Boosted trees	0.820	0.762	0.850	0.727	0.872	0.744	0.806
Test: prostate and other cancers
Classification trees	0.770	0.660	0.809	0.553	0.884	0.593	0.786
LASSSO-LR	0.776	0.773	0.783	0.514	0.921	0.621	0.778
Random forest	0.796	0.840	0.652	0.556	0.887	0.600	0.746
Boosted trees	0.786	0.840	0.609	0.538	0.875	0.571	0.724
Test: prostate cancer
Classification trees	0.622	0.550	0.661	0.353	0.818	0.435	0.648
LASSO-LR	0.667	0.667	0.667	0.400	0.857	0.500	0.667
Random forest	0.750	0.815	0.556	0.500	0.846	0.526	0.685
Boosted trees	0.750	0.778	0.667	0.500	0.875	0.571	0.722
Test: other cancers
Classification trees	0.844	0.719	0.888	0.687	0.915	0.710	0.869
LASSO-LR	0.839	0.833	0.857	0.600	0.952	0.706	0.845
Random forest	0.839	0.854	0.786	0.611	0.932	0.688	0.820
Boosted trees	0.806	0.875	0.571	0.571	0.875	0.571	0.723

Results are reported for the melanoma training data (best LOOCV estimate) and the validation data.

AUC, area under the curve; F1, F1 score; LASSO-LR, LASSO-regularized logistic regression; LOOCV, leave-one-out cross validation; NPV, negative predictive value; PPV, positive predictive value.