Dual‐Horizon Water Main Planning Using Anchored Weibull and Enhanced XGBoost Models

Water utilities must balance long‐term planning with next year's field operations. This study addresses that challenge with a dual‐horizon framework, implemented for the City of Sugar Land, Texas. The framework combines a survival model for capital planning and a machine learning classifier for operational targeting. For the capital planning, anchored Weibull curves are fitted to local break data and aligned to manufacturer life expectations, producing stable hazard estimates, and remaining‐life projections. For operations, a feature‐enhanced XGBoost model was trained on a historic asset‐year panel and validated by predicting future breaks. Its features incorporate joint type normalization, a 25 year break history, and spatial hotspot indicators. When trained through 2023 and tested on 2024, the model achieved 81.4% (area under the curve = 0.934), yielding a manageable list of 2676 pipes from 35,281 assets. Both horizons were integrated within a Geographic Information System, creating spatially clustered candidates for capital improvement programs.