david006-DS/Sme-Credit-Risk-Analysis

Machine learning model to predict Non-Performing Loans (NPLs) for SMEs in Ghana's banking sector. Includes EDA, SMOTE balancing, and comparison of Logistic Regression, SVM, and Naive Bayes models. # SME Credit Risk Analysis ## Predicting Non-Performing Loans for Ghana's Financial Sector --- ## 📋 Table of Contents 1. Project Overview 2. Business Problem 3. Dataset Description 4. Project Structure 5. Installation 6. Quick Start 7. Methodology 8. Models Implemented 9. Evaluation Metrics 10. Results Summary 11. Feature Importance 12. Business Recommendations 13. Future Improvements 14. Contributing 15. License --- ## 🎯 Project Overview This project develops a machine learning solution to predict **Non-Performing Loans (NPLs)** for Small and Medium Enterprises (SMEs) in Ghana's banking sector. The model helps financial institutions make data-driven lending decisions, reducing default rates while maintaining healthy loan portfolios. ### Key Objectives - Build a classification model to predict loan default probability - Achieve minimum **80% accuracy** and **65% recall** for NPL detection - Provide interpretable results for business stakeholders - Create a deployable model package for production use --- ## 💼 Business Problem ### Context Ghana's banking sector faces significant challenges with SME loan defaults. High NPL rates (around 20%) lead to: - **Financial losses** from unrecovered loan amounts - **Increased provisioning** requirements - **Reduced lending capacity** for the broader economy - **Higher interest rates** passed on to borrowers ### Solution Value A predictive model that accurately identifies high-risk loan applications can: | Benefit | Impact | |---------|--------| | Reduce NPL rate | Target: 20% → 15% | | Annual savings | Estimated GHS 10.7M | | Faster decisions | Automated risk scoring | | Consistent criteria | Objective assessment | --- ## 📊 Dataset Description ### Overview | Attribute | Value | |-----------|-------| | **Records** | ~5,000 loan applications | | **Features** | 25+ variables | | **Target** | `loan_status` (Performing / Non-Performing) | | **Time Period** | Historical loan data | | **Source** | Synthetic dat …