ROCAUC Ratio

Overview

Measures the ratio in ROC AUC between privileged and unprivileged groups.

Usage

Manually

Python

# Calculate difference in rocauc score

rocauc_ratio = roc_auc_ratio(df, protected_attribute, privileged_group, labels, positive_label, y_true)
print(rocauc_ratio))

Using Fairness Object

result = fo.compute(roc_auc_ratio)

Results

2.2142857135265306

These results are obtained by using the input data given in the Create Example Data page under Getting Started

Interpretation

The result is how many times greater the privileged group's ROC AUC is compared to the unprivileged group's ROC AUC. A result > 1 indicates the privileged group's ROC AUC is better than the unprivileged group's while a result < 1 indicates the unprivileged group's ROC AUC is better. A score around 1 is desirable.