yeezyyoba/ethiopian-fintech-analytics

Technical analysis and visualization of fintech data in Ethiopia. Includes automated Git setup scripts and data processing pipelines focused on financial technology growth, scalability, and market analytics. # Ethiopian Fintech Analytics Platform ### Credit Risk Modeling & Fraud Detection | End-to-End Data Science Project --- ## Overview An end-to-end data science project building a credit risk scoring and transaction fraud detection system inspired by the Ethiopian digital finance ecosystem. Covers the full pipeline from raw data to ML models, explainability, and a business-facing Power BI dashboard. --- ## Key Findings So Far - `has_delinquency` (correlation 0.3144) is the strongest default predictor - Customers with delinquency default at **22.27%** vs **2.73%** without — 8x difference - High utilization customers default at **21.08%** vs **3.79%** — 5.6x difference - Young borrowers (<30) default at **11.73%** — nearly double the overall 6.68% rate - SMOTE applied: 14:1 class imbalance fixed to 1:1 (279,862 total samples) --- ## Tech Stack | Layer | Tools | |---|---| | Data Processing | Python, pandas, NumPy, SQLite | | Machine Learning | scikit-learn, XGBoost, LightGBM | | Explainability | SHAP | | Visualization | matplotlib, seaborn, Power BI | | API | FastAPI | --- ## Project Structure --- ## Weekly Progress - [x] Week 1 — Project setup & repo structure - [x] Week 2 — Exploratory Data Analysis - [x] Week 3 & 4 — Feature Engineering & SMOTE - [ ] Week 5 — Model Training - [ ] Week 6 — Model Explainability (SHAP) - [ ] Week 7 — Power BI Dashboard - [ ] Week 8 — Final Report & API Deployment --- ## Author **Eyob Nebyou** Computer Science Student, Addis Ababa University LinkedIn | GitHub