Figure 4From: Which Is Better: Holdout or Full-Sample Classifier Design?95% bounds for true error for patient data.Back to article page