Skip to main content

Advertisement

Table 1 Noise (λ) and sensitivities (Δf) lead to ε and interpretable re-identification confidence (ρ) for k-fold adaptive composition (k=38,070)

From: The influence of differential privacy on short term electric load forecasting

λ Δf [kW] ε \(\tilde {\epsilon }_{\tilde {\delta }}\) ρ u λ
      CountingLab Lloyd Benchmark
0 - - - - -1.53 -16.81 7.80
10,000 7.57 0.00076 0.92 0.72 -17.62 -10.48 7.98
10,000 10.05 0.00100 1.23 0.77    
10,000 15.36 0.00154 1.92 0.87    
10,000 48.00 0.00480 6.46 1.00    
56,234 7.57 0.00013 0.15 0.54 -433.29 -6.49 5.94
56,234 10.05 0.00018 0.21 0.55    
56,234 15.36 0.00027 0.32 0.58    
56,234 48.00 0.00085 1.04 0.74    
100,000 7.57 0.00008 0.08 0.52 -1084.80 -2.76 3.10
100,000 10.05 0.00010 0.11 0.53    
100,000 15.36 0.00015 0.18 0.54    
100,000 48.00 0.00048 0.57 0.64    
  1. The first row states utility (u) of the hierarchical, unperturbed forecast (λ=0) over the direct forecast