mesfinmaru/Ethiopian-Household-Wealth-Prediction

This project focuses on predicting household wealth levels in Ethiopia using Machine Learning models. It analyzes socio-economic datasets to provide insights for policy-making and economic planning. ## Ethiopian Household Wealth Predictions Streamlit application and machine learning pipeline for predicting Ethiopian household wealth quintiles from five waves of World Bank LSMS-ISA / ESS survey data. ## Overview This project builds a leakage-free classification system that predicts `cons_quint` on a 1 to 5 scale, where 1 represents the poorest 20 percent of households and 5 represents the wealthiest 20 percent. The workflow combines survey decoding, cleaning, feature engineering, supervised learning, regional ranking, and an interactive Streamlit dashboard. The application is designed for academic analysis and policy-oriented exploration rather than operational targeting or allocation decisions. ## What The App Does The Streamlit interface in `app/app.py` provides the full workflow in one place: - Home page with project overview, CRISP-DM pipeline, dataset build action, and summary metrics - Data Explorer for raw previews, summary statistics, missing-value inspection, and regional comparison - EDA for distributions, bivariate analysis, correlation views, temporal trends, and shock exposure - Preprocessing audit with cleaning logs, imputation strategy, feature engineering, and preprocessing groups - Modelling workspace for training and comparing classifiers, plus per-region models - Regional Wealth Map for ranking regions and comparing two regions side by side - Predict Household form for single-household quintile prediction - About page with methods, ethics, and project references ## Core Pipeline The codebase is organized around a survey-to-model pipeline: 1. `src/sav_reader.py` decodes the Wave 2 SPSS `.sav` files in pure Python and normalizes truncated variable names. 2. `src/data_loader.py` loads each wave’s survey modules, merges them, and keeps the pipeline leakage-free. 3. `src/missing_value_handler.py` applies survey-aware imputation strategies, including wave-aware donor filling and group imputation. 4. `src/data_cleaner.py` wraps the missing- …