Fig 1.
Data resampling and evaluation procedure.
Fig 2.
Interaction of the cross-validated training and test set performances for 4 different logistic regression model types.
Scatter points were based on shuffled splits into different train-test sets. All 4 models showed anti-diagonal trend where training and test AUCs traded off against each other. Dark symbols represent previously published performances [7].
Fig 3.
Interaction of the cross-validated training and test set performances for 4 logistic regression model types across different splits into train-test sets.
Solid symbols indicate splits with significant differences in patient age or lesion size; their exclusion would not have changed the overall distributions.
Fig 4.
Interaction of the cross-validated training and test sets for two SVM model types.
Scatter points from different train-test splits are randomly distributed.
Fig 5.
Box plots of cross-validated AUCs using different numbers of training cases.