Data-driven quantification of fecal and total coliform bacteria for digital-twin-assisted water-quality monitoring

Access to safe drinking water remains a pressing global challenge, with waterborne diseases causing over 505,000 deaths annually. Total Coliform (TC) and Fecal Coliform (FC) bacteria are key indicators of water quality and safety, but traditional testing methods are time-consuming, expensive, and require specialized laboratory equipment and expertise. Recently, machine learning approaches have emerged as promising solutions for faster and more cost-effective bacterial assessments. Existing studies have addressed various aspects of microbiological water quality monitoring, including bacterial detection and concentration estimation, often focusing on specific indicators, water body types, or geographic contexts. This study proposes a novel data-driven modeling approach to enable the quantification of FC and TC bacteria across diverse types of water bodies from other easily measurable parameters. Over eight years, more than 5000 water samples were collected from 139 different locations in Andhra Pradesh, India. From the collected data, bacterial quantification models were trained. Random Forest proved to be the best-performing algorithm, with R-squared values of 0.81 and 0.74 for FCs and TCs, respectively. Furthermore, a complete digital twin system is proposed to streamline the application of the models. Thus, this study complements existing research by advancing data-driven coliform assessment and contributes to the United Nations Sustainable Development Goal 6: Clean Water and Sanitation. • A data-driven approach to the quantification of fecal and total coliform bacteria. • A comparative evaluation of six machine learning algorithms for bacterial estimation. • A SHapley Additive exPlanations (SHAP)-based feature importance analysis. • A digital twin system to streamline the application of the learning models. • Contributions to the UN Sustainable Development Goal 6: Clean Water and Sanitation.