Differential Privacy Simulator

Inject Laplace/Gaussian-style noise scaled by the privacy budget epsilon into model coefficients and trace the resulting privacy-utility frontier.

  • differential privacy
  • privacy-utility tradeoff
  • governance
  • epsilon
Mean DP AUC
0.778
SD of DP AUC
6.32e-5
True AUC (no DP)
0.778
Utility loss
-1.07e-5

AUC distribution under DP (ε = 1.0)

No DP0.30.40.50.60.70.80.9100.511.522.533.544.5AUCCount

Privacy-utility frontier

chanceNo DP limit0123456789100.40.50.60.70.80.91Privacy budget (ε)Mean AUC
There is no free lunch
Moderate privacy: AUC is degraded but usually still above chance.