Skip to main content
Account

Table 7 Summary results for Experiment 2, comparing AASPE of PAFER and random baseline with minleaf set to 0.01 under the all tested dataset and sensitive attribute combinations (setting)

From: Privacy constrained fairness estimation for decision trees

0.05

0.10

0.15

0.20

0.25

Adult::Ethnicity

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

Adult::Sex

\(p < .02\)*

\(p < .002\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

Adult::Sex-Ethn.

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

COMPAS::Ethnicity

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

COMPAS::Sex

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

COMPAS::Sex-Ethn.

\(p = .34\)

\(p < .02\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

Taiwan::Sex

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

\(p < .001\)*

  1. Adult dataset and the binary ethnicity sensitive attribute. A * indicates that PAFER performed significantly better than the random baseline (\(p < 0.05\))