⚖️ Unbiased-FairAI

Detecting and mitigating bias in ML models for fair, transparent, and explainable automated decision-making.

Google Solution Challenge 2026 | Built by Aman Kushwaha

---

🚨 The Problem

ML models silently discriminate. A model predicting loan approval or hiring decisions looks fair on the surface — but systematically disadvantages women or minorities. Nobody notices. Nobody fixes it.

Unbiased-FairAI detects and fixes exactly that.

---

🎯 What It Does

Feature	Description
📊 Pre-Model Audit	Detects bias in raw data before training
🤖 Model Training	RandomForest trained on Adult Income dataset
⚖️ Fairness Metrics	Demographic parity before & after mitigation
🔧 Bias Mitigation	ExponentiatedGradient reduces bias by 98.5%
🤖 Gemini Explainer	Plain-English explanation of WHY model is biased
🔍 What-If Simulator	Change gender/race — see if prediction changes
📋 Audit Log	Full downloadable CSV audit trail

---

📊 Results

Metric	Value
Model Accuracy	85%+
Demographic Parity Before	0.1949
Demographic Parity After	0.0029
Bias Reduction	98.5%

---

🛠️ Tech Stack

Layer	Technology
ML & Fairness	Python, Scikit-learn, Fairlearn
AI Explanation	Google Gemini API
Frontend	Streamlit
Data	UCI Adult Income Dataset

---

🚀 Getting Started

git clone https://github.com/AMAN194701/Unbiased-FairAI
cd Unbiased-FairAI
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # Add your Gemini API key
streamlit run src/app.py

---

🔑 Environment Variables

GEMINI_API_KEY=your_gemini_api_key_here

---

📄 License

MIT License