Data-Driven Poverty Profiling in Sierra Leone: An Unsupervised Machine Learning Analysis of Household Infrastructure

Background: Traditional poverty assessment in Sierra Leone often relies on singular income-based metrics, which fail to capture the complexity of household deprivation. As the nation scales its "Feed Salone" and human capital development agendas, evidence-based targeting of infrastructure support is critical. Objective: This study aims to identify and profile hidden socio-economic segments within the Sierra Leonean population by analyzing multidimensional indicators of household welfare. Methodology: Utilizing the 2019 Sierra Leone Demographic and Health Survey (DHS) Household Recode ($n=13,399$), the study implements a comparative unsupervised learning pipeline. Three variations—Principal Component Analysis (PCA), K-Means Clustering, and Agglomerative Hierarchical Clustering—were deployed to segment households based on water source, sanitation facilities, energy access, flooring material, and cooking fuel. Results: The K-Means model successfully identified four distinct pillars of poverty: the "Urban Elite" (5%), "Emerging Middle" (25%), "Rural Majority" (48%), and "Extreme Infrastructure Poverty" (22%). Findings reveal that while geography is a factor, significant "pockets of vulnerability" exist even within developed districts, particularly regarding sanitation and clean cooking fuel. Conclusion: The results provide a granular "Vulnerability Heatmap" for policymakers, suggesting that aid interventions should shift from broad district-level targeting to profile-specific infrastructure packages.