An Innovative Approach to Predict Drinking Water Risks Using System, Community, and Regulatory Characteristics.

Abstract

Robust predictive models are essential for preventing and mitigating risks associated with public drinking water systems (PWS), which pose significant public health threats and incur substantial medical costs. This study introduces a novel approach by comparing the performance of Logit, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) models in predicting risks based on PWS characteristics, community attributes, and regulatory developments, rather than relying on water quality and hydrological parameters. The study yields 3 key findings: (1) XGBoost outperforms Logit and SVM, though all models perform less effectively for predicting health-based risks; (2) community and regulatory characteristics exert a greater influence on risk predictions than PWS characteristics; and (3) XGBoost performs comparably to the water parameter-based prediction approach, with the added benefits of lower cost and suitability for long-term forecasting. This innovative approach offers substantial potential for residents, environmental advocates, and policymakers to better anticipate and address PWS risks by focusing on fundamental social determinants.