fairhealth 0.1.0

Trustworthy Healthcare AI: federated learning, fairness auditing, and explainability for clinical settings

FairHealth

Trustworthy Healthcare AI — built from peer-reviewed research.

---

FairHealth is an open-source Python library for building fair, explainable, and privacy-preserving machine learning models for healthcare.

Built by Farjana Yesmin from peer-reviewed research on maternal health, biosignals, and federated learning.

---

Install

pip install fairhealth

---

What It Does

Module	What	Paper
fairhealth.fairness	Demographic parity, equalized odds, disparate impact	MobiHealth 2026
fairhealth.explain	SHAP wrappers + Fuzzy-XGBoost hybrid explainer	ICAIHE 2026
fairhealth.federated	FedAvg + differential privacy + sparsification	MedHE, CIBB 2026
fairhealth.datasets	Maternal health, dengue, flood PDNA — all public	Multiple

---

Quick Example

import fairhealth as fh
import numpy as np

# ── Fairness audit ───────────────────────────────────────────────
from fairhealth.fairness.metrics import demographic_parity_diff

y_pred    = np.array([1, 0, 1, 0, 1, 0])
sensitive = np.array([0, 0, 0, 1, 1, 1])   # 0=young, 1=older

dpd = demographic_parity_diff(y_pred, sensitive)
print(f"Demographic Parity Difference: {dpd:.4f}")
# → 0.3333  (gap between age groups)

# ── Fuzzy risk explanation ───────────────────────────────────────
from fairhealth.explain.fuzzy import get_fired_rules, score_to_label

rules = get_fired_rules(age=42, sbp=145, bs=12, hr=88)
for rule in rules:
    print(f"Rule {rule['id']}: {rule['condition']} → {rule['outcome']}")
# → Rule 1: High BP AND High Blood Sugar → HIGH RISK
# → Rule 5: High Heart Rate AND High BP  → HIGH RISK

# ── Federated privacy ────────────────────────────────────────────
from fairhealth.federated.privacy import sparsify, add_gaussian_noise

weights       = np.array([0.4, 0.3, 0.15, 0.1, 0.03, 0.02])
sparse, rate  = sparsify(weights, sparsity=0.975)
print(f"Communication reduced by {rate:.1%}")
# → Communication reduced by 83.3%

---

Real Results (from My Papers)

Finding	Value
Maternal health model accuracy	79.3% (Fuzzy-XGBoost hybrid)
Clinicians preferring hybrid explanation	71% (14 clinicians, ICAIHE 2026)
Demographic parity difference (age groups)	0.1011
Federated vs central accuracy gap	9.3%
Communication reduction (sparsification)	83–97.5%

---

Datasets Used (All Public — No Hospital Access Needed)

Dataset	Domain	Source
Maternal Health Risk	Risk prediction	UCI / Kaggle
Bangladesh Dengue	Symptom triage	DGHS Bangladesh
Bangladesh Flood PDNA 2022	Disaster equity	Government open data
PTB-XL	ECG biosignals	PhysioNet (free)

---

Research Papers

If my library helps your work, please cite:

@software{fairhealth2026,
  author = {Yesmin, Farjana},
  title  = {FairHealth: Trustworthy Healthcare AI},
  year   = {2026},
  url    = {https://github.com/Farjana-Yesmin/fairhealth}
}

Related papers:

• Yesmin, F. (2026). Fairness-Aware Representation Learning for ECG-Based Disease Prediction. MobiHealth 2026.
• Yesmin, F. et al. (2026). Explainable AI for Maternal Health Risk Prediction in Bangladesh. ICAIHE 2026.
• Yesmin, F. (2026). MedHE: Communication-Efficient Privacy-Preserving Federated Learning. CIBB 2026.
• Yesmin, F. & Akter, R. (2026). Toward Equitable Recovery. CCAI 2026 (IEEE).

---

Author

Farjana Yesmin — ML Researcher, Trustworthy AI for Healthcare
Website: farjana-yesmin.github.io
Email: [e-mail]

---

License

MIT © Farjana Yesmin