Empowering Financial Inclusion with AI-Driven Credit Risk Assessment

• Introduction
• ✨ Features
• 📂 Project Structure
• ⚙️ Installation
• 🚀 Usage
  • Running the Backend
  • Using the Frontend
  • Making API Predictions
• 🔍 Exploratory Data Analysis (EDA)
• 🛠️ Feature Engineering
• 🤖 Model Training and Evaluation
• 🔮 Model Explainability
• 🌐 API Development
• 💻 Frontend Interface
• 🚀 Deployment
• 🎯 Challenges and Solutions
• 🔮 Future Work
• 🤝 Contributing

The Credit Scoring Model for Bati Bank is an AI-powered platform designed to assess credit risk using eCommerce transaction data. This solution enables financial inclusion through:

• 📈 Accurate Predictions: Random Forest model achieves ROC-AUC: 0.9998
• 🔍 Transparent Decisions: SHAP explanations and feature importance visualizations
• ⚡ Real-Time Processing: FastAPI backend with <100ms response times
• 📱 Mobile-First Interface: Responsive design accessible on all devices

• Automated Data Pipelines
  • RFMS scoring (Recency, Frequency, Monetary, Score)
  • WoE encoding for categorical features
• Advanced Modeling
  • Hyperparameter-tuned Random Forest & Logistic Regression
  • Cross-validation with stratified sampling
• Production-Ready Deployment
  • Dockerized environment
  • CI/CD pipeline with GitHub Actions
• User-Centric Interface
  • Dual form system (Quick/Detailed assessment)
  • Interactive risk visualization dashboard

dagiteferi-credit-scoring-model/
├── 📁 credit_scoring_app/       # FastAPI backend
├── 📁 models/                   # Serialized ML models
├── 📁 notebooks/                # Jupyter analysis notebooks
├── 📁 scripts/                  # Data processing scripts
├── 📁 static/                   # CSS/JS assets
└── 📁 tests/                    # Unit/integration tests

git clone https://github.com/your-repo/dagiteferi-credit-scoring-model.git
cd dagiteferi-credit-scoring-model
python3 -m venv venv
source venv/bin/activate  # Linux/Mac
venv\Scripts\activate     # Windows
pip install -r requirements.txt

## 🚀 Usage

### Running the Backend
```bash
cd credit_scoring_app
uvicorn main:app --host 0.0.0.0 --port 8000

Access at http://localhost:8000/static/index.html

curl -X POST "http://localhost:8000/predict/good" \
-H "Content-Type: application/json" \
-d '{
  "TransactionId": 1,
  "Amount": 0.05,
  "FraudResult": 0
}'

Key Insights:

• 🎯 Class Imbalance: Only 0.2% fraud cases
• 📉 Skewed Distributions: Transaction amounts follow power law
• 🔗 Strong Correlations:
  • RFMS_score ↔ Total_Transaction_Amount (ρ=0.89)
  • Transaction_Count ↔ Product_Variety (ρ=0.76)

Transformations Applied:

1. Temporal Features
   • Transaction hour/day/month
   • Time since last transaction
2. Aggregate Features
   • 30-day rolling transaction count
   • Customer lifetime value

SHAP Analysis:

• Top Predictive Features:
  1. Total_Transaction_Amount (SHAP value: 1.42)
  2. RFMS_score (SHAP value: 1.18)
  3. Transaction_Recency (SHAP value: 0.76)

Endpoints:

@app.post("/predict/good")
async def predict_good_risk(data: CustomerData):
    return predict(data, model_path="models/RandomForest_best_model.pkl")