Uncovering Hidden Water Insecurity: A Conflict‐Augmented, Equity‐Weighted Machine Learning Framework for Global SDG 6.1 Risk Assessment

ABSTRACT Achieving Sustainable Development Goal 6.1—universal safe drinking water access by 2030—demands forward‐looking risk intelligence that moves beyond trend extrapolation, yet monitoring frameworks report current coverage and cannot anticipate its deterioration. We merge WHO/UNICEF Joint Monitoring Programme, World Bank, ACLED conflict, ND‐GAIN climate vulnerability and Fragile States Index data into a 232‐country panel spanning 2000–2024, and predict from 15 features—eight socioeconomic and service‐coverage, seven conflict‐climate—whether safely managed coverage will fall below 50% 5 years ahead ( country‐years). Under country‐grouped cross‐validation a persistence baseline dominates rank‐order discrimination (AUC = 0.968), a finding that should reframe how such models are judged: machine learning contributes through calibration, equity and attribution rather than ranking. Conflict and climate covariates improve calibration (Brier 0.190 vs. 0.196) and yield a Hidden Vulnerability Index, evaluated against seed, architecture and permutation controls, that ranks countries whose reported trajectories understate structural risk. Equity weighting narrows the equal‐opportunity gap across income groups by 41.5%, and temporal attribution identifies the urban–rural gap as the fastest‐rising predictor of future insecurity ( pp/year). Only 40.5% of countries are projected to achieve SDG 6.1, with no low‐income country on track.