Paper Title
A Data-Intelligence Framework using Machine Learning to Detect and Map Health Deserts in India

Abstract
Ensuring equitable access to primary healthcare is a critical challenge in India, yet national-scale assessments are often hindered by data noise and computational limits. We introduce a reproducible data-intelligence framework that integrates World Pop 1-km population grids, NFHS-5 district covariates, and a cleaned national facility registry (n = 200,421) to identify "health deserts." Using a grid-based Two-Step Floating Catchment Area (2SFCA) method, we computed population-weighted accessibility scores for 676 districts. Contiguous low-access zones were delineated using density-based spatial clustering (DBSCAN). To operationalize prioritization, we trained Random Forest and XGBoost classifiers to predict desert status, achieving a PR-AUC of 0.758. Crucially, we applied SHAP (Shapley Additive Explanations) to interpret model outputs, identifying maternal health infrastructure and sanitation deficits as key drivers of inaccessibility. This framework provides policymakers with validated, spatially coherent priority zones for targeted intervention. Keywords - Health Deserts, Spatial Accessibility, 2SFCA, XGBoost, SHAP, Geospatial Health