Privacy Metrics for Synthetic Data

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.

Distance and Similarity 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

Attack-Based Metrics

Metrics supported for attribute disclosure as ML Task:

• Accuracy and F1 Score for Classification Tasks
• MAE, R squared and MAPE for Regression Tasks