scikit-fair¶

Fairness-aware machine learning toolkit with a scikit-learn compatible API.

scikit-fair (skfair) is a Python library for fairness-aware binary classification. It covers the full pipeline — preprocessing, evaluation, auditing, comparison, and experimentation — and integrates seamlessly with scikit-learn and imbalanced-learn workflows.

What it does¶

Algorithmic fairness is concerned with preventing machine learning models from discriminating against individuals based on sensitive attributes such as race, sex, or age. scikit-fair provides tools for every stage of the fairness workflow:

Preprocessing: transform data before training — weighting, resampling, and feature transformation techniques
Metrics: nine group-fairness metrics and nine performance metrics with a unified API
Datasets: five standard fairness benchmark datasets with convenient loaders
Audit: pre-model data analysis (BiasAuditor) and post-model fairness evaluation (FairnessAuditor)
Comparison: visual comparison of multiple preprocessing methods across datasets and classifiers (ComparisonReport)
Experimentation: automated dataset × method × classifier experiments with cross-validation (Experiment)

At a glance¶

from skfair.preprocessing import Massaging
from skfair.metrics import disparate_impact, statistical_parity_difference
from skfair.datasets import load_adult

from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split

X, y = load_adult(return_X_y=True, as_frame=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Reduce discrimination via label massaging
sampler = Massaging(sens_attr="sex", priv_group=1)
X_fair, y_fair = sampler.fit_resample(X_train, y_train)

clf = LogisticRegression(max_iter=1000)
clf.fit(X_fair, y_fair)
y_pred = clf.predict(X_test)

# Measure improvement
print(disparate_impact(y_test.values, y_pred, X_test["sex"].values))

Algorithms¶

Class	Type	Reference
`Reweighing`	Weighting	Kamiran & Calders (2012)
`FairBalance`	Weighting	Yu et al. (2024)
`ReweighingClassifier`	Meta-estimator	—
`FairBalanceClassifier`	Meta-estimator	—
`Massaging`	Label modification	Kamiran & Calders (2012)
`FairwayRemover`	Undersampling	Fairway (2019)
`FairOversampling`	Oversampling	Dablan et al.
`FairSmote`	Oversampling	Chakraborty et al. (2021)
`FAWOS`	Oversampling	Salazar et al. (2021)
`HeterogeneousFOS`	Oversampling	Sonoda et al. (2023)
`DisparateImpactRemover`	Feature transformation	Feldman et al. (2015)
`OptimizedPreprocessing`	Feature transformation	Calmon et al. (2017)
`LearningFairRepresentations`	Feature transformation	Zemel et al. (2013)
`FairMask`	Meta-estimator	Peng et al. (2021)
`IntersectionalBinarizer`	Utility	—
`DropColumns`	Utility	—

Metrics¶

Nine group fairness metrics and nine performance metrics are included. All follow a consistent signature: metric(y_true, y_pred, sensitive_attr).

Example notebooks¶

The examples/ folder contains step-by-step Jupyter notebooks that walk through every module: