thandofana/finaccess-eswatini

Explainable machine learning system for financial inclusion and mobile-money adoption in Eswatini. # FinAccess Eswatini An explainable machine-learning system that analyses financial access in Eswatini and estimates whether an individual profile is likely to be financially included or use mobile money. ## 1. Project Overview Financial inclusion affects how people save, receive payments, manage emergencies, and participate in the wider economy. However, access to formal financial services and mobile money is not distributed equally. FinAccess Eswatini uses World Bank Global Findex microdata to answer two questions: - Is a person with a given demographic, socioeconomic, and digital-access profile likely to be financially included? - Is that person likely to use mobile money? The project combines survey-weighted analysis, statistical testing, two independently designed machine-learning models, SHAP explanations, a FastAPI service, and a responsive web application. Potential users include researchers, financial-inclusion practitioners, policymakers, and development organisations. It is a proof of concept and not a financial eligibility or credit decision system. ## 2. Project Objectives - Measure and describe financial inclusion and mobile-money adoption in the Eswatini dataset. - Identify demographic, socioeconomic, and digital characteristics associated with both outcomes without making causal claims. - Build separate, leakage-reviewed prediction pipelines for financial inclusion and mobile-money adoption. - Explain global model behaviour and individual predictions using SHAP. - Deliver both models through one validated API and one accessible web assessment. ## 3. Dataset | Item | Description | |---|---| | Source | World Bank Global Findex Eswatini microdata | | Survey reference | `SWZ_2024_FINDEX_v02_M` | | Publication/database edition | Global Findex 2025 | | Respondents | 1,051 | | Raw variables | 199 | | Model 1 target | `account_fin` - financial institution account ownership | | Model 2 target | `account_mob` - mobile-money account ownership | | Mod …