Problem
Formal credit scoring systems rely on bank account history, loan repayment records, and formal employment — data that most Malawians don't have. This excludes the majority from accessing loans and financial products.
Solution
CreditIQ ingests alternative data sources: mobile money transaction frequency, airtime purchase patterns, utility payment regularity, and social network signals. A gradient-boosted model trained on synthetic + real-world proxy data assigns risk tiers. The model is served via a FastAPI REST endpoint with JSON input/output, containerized with Docker.
Real-World Impact
Demonstrates a pathway for microfinance institutions and fintech startups in Malawi and Sub-Saharan Africa to extend credit responsibly to informal economy workers.
Challenges Faced
Building a fair, explainable model without reinforcing socioeconomic bias. Used SHAP values to audit feature importance and ensure the model doesn't penalize poverty proxies.
Key Learnings
Credit scoring is as much an ethics problem as a technical one. SHAP explainability wasn't optional — it was essential for building trust in the model's outputs.
Demo & Execution Screenshots
