This repository implements a set of privacy metrics proposed in the literature for measuring disclosure risks in synthetic data.
We identify two main groups of metrics: Distance and Similarity metrics and Attack-Based Metrics.
The metrics covered are the following:
- Exact Matches
- Distance to Closest Record (DCR)
- Nearest Neighbor Distance Ratio (NNDR)
- Outliers Similarity
- Cosine Similarity
- Hausdorff Distance
Metrics supported for attribute disclosure as ML Task:
- Accuracy and F1 Score for Classification Tasks
- MAE, R squared and MAPE for Regression Tasks