🚀 CrediSense AI

An advanced machine learning system for predicting loan default risk in microfinance institutions, tailored for African financial ecosystems. A scalable, explainable AI system for predicting loan default risk and optimizing lending decisions in microfinance environments like Susu Digital.

---

🧠 Overview

CrediSense AI is an advanced machine learning system designed to predict loan default risk in microfinance environments.

Built for platforms like Susu Digital, it enables:

• Data-driven lending decisions
• Risk-aware credit scoring
• Explainable AI insights

---

🎯 Key Features

• XGBoost model with hyperparameter tuning
• 1M+ synthetic financial dataset
• SHAP explainability (global + local)
• Risk threshold optimization
• REST API with FastAPI
• Production-ready architecture

---

📊 Model Capabilities

• Predict default probability
• Classify risk (Low / Medium / High)
• Optimize decision thresholds
• Explain predictions using SHAP

---

📊 Dataset

Sources:

• Synthetic dataset (generated)

Features:

• Age
• Income
• Loan Amount
• Savings Balance
• Missed Payments
• Loan Duration
• Debt-to-Income Ratio
• Savings Ratio
• Payment Stress

---

⚙️ Tech Stack

• Python
• Pandas / NumPy
• Scikit-learn
• XGBoost
• FastAPI
• Matplotlib
• SHAP

---

🚀 Getting Started

1. Install dependencies

pip install -r requirements.txt

2. Generate dataset

python src/generate_data.py

3. Train model

python src/train.py

4. Run API

cd api
uvicorn main:app --reload

5. API Testing

curl -X POST http://127.0.0.1:8000/predict \
-H "Content-Type: application/json" \
-d @tests/test.json

---

📈 Risk Threshold Optimization

Instead of fixed 0.5 threshold:

• Model finds optimal threshold using Precision-Recall tradeoff
• Improves recall (catching defaulters)
• Reduces financial risk

---

🔍 Explainability (SHAP)

• Feature importance visualization
• Per-user explanation
• Transparent decision-making

---

📈Model Accuracy

Metrics used:

Accuracy & Precision

ROC-AUC (primary metric)

🔮 Future Improvements

Deep learning (LSTM for time-series financial behavior)

Real-time streaming predictions

Integration with mobile money APIs

React dashboard for loan officers

🌍 Vision

To build an AI-powered credit scoring engine for Africa, enabling:

Financial inclusion

Risk reduction

Data-driven lending

🤝 Contribution Guide

• Fork the repo
• Create feature branch
• Commit changes
• Submit PR

---

👨‍💻 Author

Boateng Prince Agyenim (Mmabiaa) Fullstack Engineer | AI Developer Founder, Susu Digital