Takunda-Madondo/loan-intelligence

Loan Intelligence. XGBoost loan default prediction system for African microfinance markets, with a Streamlit dashboard for portfolio risk analysis, investment signals, and AI-powered querying via Claude API. # 🏦 Loan Intelligence & Portfolio Optimization ### Predicting loan defaults and identifying high-return sectors in African microfinance markets --- ## What This Project Does Most African microfinance borrowers have no formal credit history. This system solves three problems simultaneously: 1. **Default Risk** — Which borrowers are likely to default, and why? 2. **Capital Allocation** — Which sectors (agriculture, retail, services) deliver the best return on lending? This project builds a full data science solution — from raw data ingestion to a live dashboard — that answers both questions using real-world loan data. **Target Users:** Credit risk teams, loan officers, MFI portfolio managers --- ## Dashboard Pages | Page | Business Question | |---|---| | Portfolio Overview | What is our overall portfolio health and model performance? | | Investment Signals | Which sectors should we increase or reduce exposure to? | | Portfolio Optimisation | Given our budget and risk philosophy, what is the optimal capital allocation? | | Default Risk Map | Which African markets carry the most systemic risk? | | Loan Predictor | What is the default risk on this specific loan application? | Every page includes an **Ask AI** panel powered by the Anthropic Claude API for natural language querying of portfolio data. --- ## Architecture ``` Raw Data (3 sources) │ ▼ Bronze Layer ── bronze.db (raw ingestion) kiva_loans_raw · worldbank_raw · lending_club_raw │ ▼ Silver Layer ── silver.db (clean + join) lc_clean · kiva_clean · macro_clean · kiva_enriched │ ▼ Gold Layer ── gold.db (ML-ready) lc_features · kiva_features · sector_performance kiva_predictions · portfolio_allocations │ ▼ Dashboard ── Streamlit (5 pages) ``` --- ## Key Technical Decisions **Transfer learning** — Kiva's public dataset has no repayment outcomes. The model trains on Lending Club's verified labels and scores Kiva loans using shared features (loan amount, term, sector, repayment …