BruceOnyango/ea-financial-inclusion

End-to-end ML predicting bank-account ownership across Kenya, Rwanda, Tanzania and Uganda from FinScope survey data. Five classifiers compared with leakage-safe 2-fold CV, SMOTE for imbalance, thresholds tuned out-of-fold. Winning SVM served through an interactive Dash dashboard that maps exclusion and screens individuals. 100% test coverage. # Financial Inclusion in East Africa Across Kenya, Rwanda, Tanzania and Uganda, only about **14 in every 100 adults hold a bank account**. This project uses survey data to predict who is likely to be excluded, shows where the gaps are widest, and serves a live screener through an interactive Dash dashboard. The emphasis is a defensible, end to end machine learning workflow: from data construction through leakage safe cross validation, honest evaluation under class imbalance, and a dashboard where every number on screen is one a unit test verified. > **Headline result.** Five classifiers reach near identical accuracy (around 85%), > but accuracy is misleading when only 14% of people are banked. Judged on the metrics > that matter for finding the excluded, an **SVM (RBF)** leads, correctly ranking a > banked and an unbanked person about 86 times in 100 (ROC-AUC 0.858) and recovering > 64% of truly banked adults. --- ## Table of contents - The problem - Data - Quickstart - Reproducing the model - The dashboard - Testing and coverage - Project structure - Methodology - Key findings --- ## The problem Access to a bank account lets households save, make payments, and build the credit history that unlocks further finance. It is a recognised contributor to long term economic growth. Yet across these four East African countries the majority of adults remain unbanked. Knowing **who** is excluded, and **which factors** drive that exclusion, turns a broad policy goal into targeted, measurable action for a bank, an NGO, or a financial regulator. This is framed as a binary classification task: given a person's demographic profile, predict whether they hold a bank account (Yes = 1, No = 0). --- ## Data | Field | Detail | | --- | --- | | Source | FinScope and FinAccess household surveys, 2016 to 2018 | | Coverage | Kenya, Rwanda, Tanzania, Uganda | | Rows | 23,524 labelled respondents | | Target | `bank_account` (14.1% positive) | | Features | 10 demographic attri …