Fig. 9From: A comprehensive tool for creating and evaluating privacy-preserving biomedical prediction modelsROC performance in the case study using differential privacy for training logistic regression models to predict the malignancy of breast tissue. The False Positive Rates (FPR) and True Positive Rates (TPR) are plotted against the x-axes and y-axes, respectively. It can be seen that data anonymization had a significant impact on prediction performance, but acceptable accuracy could still be observed for ε≥1Back to article page