Figure 4: Relative density of classifier true positive (TP) and false positive (FP) detections on the validation set by submodel and training pass. “Final, Ensemble” (lower right) was the model used to classify survey data.