Peterson-Muriuki/msme-credit-scoring

AI-powered credit risk assessment for African MSMEs using alternative data and Smolagents # MSME Credit Risk Scoring System An AI-powered credit risk assessment system for Micro, Small, and Medium Enterprises (MSMEs) in Africa, enhanced with **Smolagents** for intelligent insights. ## Project Overview This system addresses the **$360B financing gap** for African MSMEs by leveraging alternative data sources and machine learning to assess creditworthiness beyond traditional methods. ### Key Features - **97.5% Repayment Rate** (Target: >95%) - **2.5% Default Rate** (Target: 95% - Default Rate < 3% ## Key Features by Importance Top 10 predictive features: 1. **Previous Default** (15%) - Strong negative signal 2. **Payment Score** (12%) - Reliability indicator 3. **Business Age** (10%) - Experience matters 4. **Debt-to-Income** (9%) - Capacity assessment 5. **MM Tenure** (8%) - Transaction history 6. **Social Score** (7%) - Network trust 7. **Transaction Volume** (7%) - Business activity 8. **Business Permit** (6%) - Formalization 9. **Loan Amount** (5%) - Exposure size 10. **Country Risk** (4%) - Geographic factor ## Streamlit Dashboard ### Features: 1. **Risk Assessment Tab** - Real-time default probability - Risk classification - Approval recommendations - Key risk/protective factors 2. **Portfolio Analytics** - Financial metrics - Business maturity indicators - Payment behavior analysis 3. **AI Insights** (Smolagents) - Intelligent feature analysis - Personalized recommendations - Model improvement suggestions 4. **About** - System documentation - Performance metrics - Technology stack ## Deployment ### Deploy to Streamlit Cloud ## Documentation - **Feature Engineering**: 43+ derived features from alternative data - **Model Training**: XGBoost, Logistic Regression with SMOTE - **Evaluation**: ROC curves, threshold optimization, portfolio analysis - **Smolagents**: AI-powered insights and recommendations ## Quarterly Updates The system supports quarterly model iterations: 1. Collect new loan data 2. Run feature engineer …