Machine Learning Approaches to Predicting Poverty and Welfare Outcomes from Demographic Data in Nigeria

Poverty measurement in developing countries is hindered by costly, infrequent traditional surveys and the limitations of conventional econometric models, motivating the use of transparent and interpretable machine-learning approaches. This study examines application of machine learning (ML) techniques in predicting poverty and welfare outcomes in Nigeria using demographic and socio-economic variables, routinely available at scale. Traditional poverty measurement approaches, primarily consumption surveys remain costly, not frequent, and spatially limited, resulting in substantial delays in policy targeting. The study develops and evaluates a suite of ML models, including Logistic Regression, Random Forest and Gradient Boosting, trained on 50,000 representative household data. These models generate household-level poverty probabilities and identify the key predictors of welfare deprivation. Empirical results show that ensemble models, particularly Random Forest and Gradient Boosting, outperform linear models, achieving higher predictive accuracy and discriminatory power (AUC > 0.90). Analyses highlight education level, household size, rural residence, access to electricity, employment status, and asset endowment as the strongest predictors of poverty. Shapley Additive exPlanations (SHAP) and partial dependence plots reveal non-linear relationships that traditional econometric methods often fail to capture. The study also produces Receiver Operating Characteristic (ROC) curves, predicted probability distributions, and visualization of output to support model evaluation. Policy simulations demonstrate that integrating ML based targeting frameworks into Nigeria’s social protection systems can significantly reduce exclusion and inclusion error, improving the efficiency of cash transfers and welfare programmes. It concludes that ML approaches offer a cost-effective and scalable complement to conventional poverty measurement. scalable complement to conventional poverty measurement.