Figure 2. Model metrics obtained from training the random forest model across a range of cutoffs from 0.01 to 0.99 in increments of 0.01.

To train the random forest model we used leave-one-out cross-validation across a range of cutoffs from 0.01 to 0.99 in increments of 0.01. Raising the cutoff value means a higher level of evidence is needed to assign the positive class (uncontacted), which decreases sensitivity (true positive rate) and increases specificity (true negative rate). Here the optimal cutoff (0.2) gives a perfect cross-validated sensitivity of 1.0 and a specificity of 0.98. The distance is the distance from a perfect model which is minimized during training.