mercy-ndungu/kenya-health-fairness-audit

# Kenya Health Infrastructure Analysis ## A Human-Centered AI Research Project ### Project Overview 97% of pharmacies in Kenya have no verified registration number on record. 10 counties serving 8 million people fall below the WHO minimum for hospital beds. This project asks whether those patterns reflect current population needs or decisions made decades ago that no one has revisited. It then asks what happens when you build a machine learning model on data that contains those patterns. ### The Research Question Does geographic accessibility to health facilities in Kenya reflect population needs, or historical infrastructure bias? This is not a question about bad data. It is a question about what data cannot see, the unregistered clinics, the community health workers, the informal infrastructure that concentrates in exactly the counties where the formal system is thinnest. It is also a question of what machine learning models learn when trained on records that reflect decades of unequal investment. This project addresses that question in four phases: cleaning and understanding the raw data, building a county-level picture of health infrastructure across all 47 counties, training predictive models, and assessing whether those models are fair, explainable, and safe to use for decisions that affect real people. *Figure 1. Hospital beds per county in Kenya. Red = under 1,000 beds. Orange = 1,000–2,000. Green = over 2,000. The dashed line is the national average. Ten counties fall below the WHO minimum of 10 beds per 10,000 people.* ### Findings 97% of pharmacies have no registration number on record. Pharmaceutical regulation is effectively unverifiable at the county level. 10 counties fall below the WHO minimum of 10 beds per 10,000 people. 8 million people live in counties with inadequate inpatient capacity33.7% of health units are not fully operational. Facility counts overstate real capacity. Semi-functional facilities look like full ones. The predictive …