Table 2

Performance measures for the four classifiers

AccuracySensitivitySpecificityPPVNPVF1AUC
Training: melanoma (LOOCV estimate)
 Classification trees0.7940.8400.7740.6670.9140.7530.748
 LASSO-LR0.8030.9050.7500.6550.9380.7600.827
 Random forest0.7540.7620.7500.6150.8570.6810.756
 Boosted trees0.8200.7620.8500.7270.8720.7440.806
Test: prostate and other cancers
 Classification trees0.7700.6600.8090.5530.8840.5930.786
 LASSSO-LR0.7760.7730.7830.5140.9210.6210.778
 Random forest0.7960.8400.6520.5560.8870.6000.746
 Boosted trees0.7860.8400.6090.5380.8750.5710.724
Test: prostate cancer
 Classification trees0.6220.5500.6610.3530.8180.4350.648
 LASSO-LR0.6670.6670.6670.4000.8570.5000.667
 Random forest0.7500.8150.5560.5000.8460.5260.685
 Boosted trees0.7500.7780.6670.5000.8750.5710.722
Test: other cancers
 Classification trees0.8440.7190.8880.6870.9150.7100.869
 LASSO-LR0.8390.8330.8570.6000.9520.7060.845
 Random forest0.8390.8540.7860.6110.9320.6880.820
 Boosted trees0.8060.8750.5710.5710.8750.5710.723
  • 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.