Fig. 4.

Different approaches for measuring the relative performance of a privacy-preserving classifier. Our tool implements a method that makes sure that the performance of prediction models can be expressed relative to the performance of models trained on unmodified data