---

What is FairCheck?

FairCheck is an open-source Python library that makes fairness analysis in machine learning simple, accessible, and actionable. Whether you're a researcher, ML engineer, or AI ethics practitioner, FairCheck helps you understand how your models perform across demographic groups.

Why FairCheck?

Modern AI systems often exhibit demographic biases that are invisible in aggregate metrics. A model with 95% accuracy might perform dramatically worse for certain groups. FairCheck makes detecting and mitigating these biases as easy as one function call.

from faircheck import audit

results = audit(
    model=your_model,
    dataset=your_data,
    sensitive_attrs=['gender', 'race', 'age']
)

results.report()  # Beautiful interactive dashboard

That's it. No more 100-line scripts. No more custom analysis pipelines.

Key Features

• One-line auditing - Comprehensive fairness analysis in a single function call
• Multiple metrics - Demographic parity, equalized odds, TPR/FPR disparity, calibration, and more
• Framework agnostic - Works with PyTorch, TensorFlow, Scikit-learn, and Hugging Face models
• Beautiful reports - Interactive HTML dashboards, PDF exports, and JSON outputs
• Bias mitigation - State-of-the-art techniques to reduce bias in trained models
• Continuous monitoring - Track fairness in production over time
• Research-grade - Implements peer-reviewed methods from leading fairness research

Installation

Note: FairCheck is currently in active development. The PyPI release is coming soon!

# Coming soon
pip install faircheck

# Development version
git clone https://github.com/kasra-kakavand/faircheck.git
cd faircheck
pip install -e .

Quick Start

Basic Audit

import faircheck as fc

# Audit any classifier in one line
results = fc.audit(
    model=my_model,
    dataset=test_data,
    sensitive_attrs=['gender']
)

# View results
print(results.summary())

# Generate detailed report
results.report(output='fairness_report.html')

Bias Mitigation

from faircheck import mitigate

# Apply variance-based fairness regularization
fair_model = mitigate(
    model=biased_model,
    method='variance_regularization',
    train_data=training_data,
    sensitive_attr='skin_tone'
)

Continuous Monitoring

from faircheck import FairnessMonitor

# Set up production monitoring
monitor = FairnessMonitor(model=production_model)

# Track predictions over time
monitor.track(predictions, demographics)

# Get alerts when fairness drifts
if monitor.detect_drift():
    monitor.alert("Fairness threshold violated")

Supported Metrics

FairCheck implements a comprehensive suite of fairness metrics:

Category	Metrics
Group Performance	Accuracy, TPR, FPR, TNR, FNR per group
Disparity Measures	Demographic parity, equalized odds, equal opportunity
Statistical	Disparate impact ratio, statistical parity difference
Calibration	Calibration error per group, ECE disparity
Custom	Variance-based regularization, intersectional metrics

Why This Matters

"Fairness in AI is not a feature—it's a requirement."

As AI systems are deployed in healthcare, hiring, criminal justice, and finance, ensuring equitable treatment across demographic groups is critical. FairCheck democratizes fairness analysis, making it accessible to every ML practitioner.

Roadmap

• Project foundation
• Core fairness metrics module
• PyTorch integration
• Interactive HTML reports
• PyPI release (v0.1.0)
• TensorFlow integration
• Bias mitigation algorithms
• LLM fairness analysis
• Production monitoring tools
• Hugging Face Hub integration
• Comprehensive documentation site

Contributing

We welcome contributions! FairCheck is built by the community, for the community. Whether you're fixing bugs, adding features, or improving documentation, your help is appreciated.

# Fork the repo, then:
git clone https://github.com/YOUR_USERNAME/faircheck.git
cd faircheck
pip install -e ".[dev]"
pytest

Citation

If you use FairCheck in your research, please cite:

@software{kakavand2026faircheck,
  title={FairCheck: A Library for Fairness Auditing in Machine Learning},
  author={Kakavand, Kasra},
  year={2026},
  url={https://github.com/kasra-kakavand/faircheck}
}

License

FairCheck is released under the MIT License. See LICENSE for details.

Author

Kasra Kakavand

• GitHub: @kasra-kakavand
• Research: Fairness-Aware Deepfake Detection

Acknowledgments

FairCheck builds on decades of research in algorithmic fairness. Special thanks to the contributors of fairness research and the open-source ML community.

---