Author(s): Killian Gleeson; Grigorios Kyritsakas; Stewart Husband; Joby Boxall
Linked Author(s):
Keywords: Drinking water distribution systems; Explainable AI; SHAP analysis; Water quality
Abstract: Water utilities struggle to extract actionable insights from their historic datasets. This study applies explainable AI to identify key drivers of water quality in distribution systems. 169,000 customer tap samples from a UK utility (2009-2022) were analysed, linking them to treatment works data and network metrics. Machine learning models were developed for free chlorine and total iron, achieving Matthews correlation coefficients of 0.49 and 0.41 respectively. SHapley Additive exPlanations analysis revealed that treatment works parameters dominate chlorine and iron predictions. This explainable AI approach transforms historic data into parameter-specific insights, enabling targeted interventions.
Year: 2026