Predictive models for public health: leveraging machine learning to analyze health inequalities

This thesis explores the feasibility of deploying machine-learning models as decision-support tools for public-health governance. Using the 2009 UK wave of the EU-SILC survey, a multidimensional dataset (≈13 k adults; >150 socio-economic, demographic and environmental proxies) was engineered into an analy­sis-ready matrix and a composite binary indicator of ill-health. A suite of supervised algorithms—logistic and gradient-boosting trees, random forests and support-vector machines—was trained and cross-validated to predict adverse health status. Explainable-AI techniques were embedded to quantify feature influence, consistently foregrounding age, income, educational attainment, housing conditions and social participation. The resulting models achieved solid discrimination and calibration, illustrating how integrated, interpretable analytics can identify vulnerable sub-populations and inform equitable, resource-efficient public-health strategies.