kebochig/Credit-Scoring-System-for-African-Market

A comprehensive machine learning system for African credit scoring with production-ready API deployment, optimized model training, monitoring, and persistence for audit. # Credit Scoring System A comprehensive machine learning system for African credit scoring with production-ready API deployment, optimized model training, monitoring, and persistence for audit. ## 📋 Overview This system provides an end-to-end solution for credit scoring in African markets, featuring: - **Hybrid Data Strategy**: Real Lending Club credit data (50K+ records) + synthetic mobile money features - **Advanced ML Models**: Optimized XGBoost with SMOTEENN (AUC-ROC ~0.77) - **FastAPI REST API**: Production-ready API with comprehensive documentation - **Real-time Scoring**: Sub-second prediction response times - **Explainable AI**: SHAP-based decision explanations for regulatory compliance - **Health Monitoring**: Service health checks and performance metrics - **Request Tracking**: Request ID tracking and error handling ## 🏗 Architecture ### 1. System Components (Pictorial Flow) ``` ├── Credit Scoring System │ ├── Data Pipeline (ETL) │ ├── ML Training & Optimization │ ├── Model Deployment (API) │ ├── Monitoring & Alerting │ └── Data Persistence ``` ### 2. Request Logic (Process Flow) 1. **Request Ingestion**: Client sends applicant data via `POST /predict`. 2. **Feature Pipeline**: The `FeatureEngineer` transforms raw data into ML-ready vectors. 3. **Inference**: The `XGBoost` model computes the default probability and risk class. 4. **Explanation**: `SHAP` values are generated to explain the top driving factors of the score. 5. **Persistence**: The request and result are logged to `SQLite` via an async background task. 6. **Monitoring**: Real-time drift detection calculates Z-scores for inputs and distribution shifts for outputs. 7. **Response**: Final score, risk level, and behavioral explanations are returned to the client. ## 🚀 Quick Start ### Prerequisites - **Python**: 3.9 or higher - **Docker**: Latest version (for containerized deployment) - **Git**: For cloning the repository ### 1. Clone Repository ```bash git clone https:/ …