Risk assessment of water distribution networks through an integrated model based on machine learning and statistical methods

Water distribution networks (WDNs) are critical for urban infrastructure, but as they expand and age, the risk of pipeline ruptures and leaks grows. Predicting these risks is essential for preventing accidents, improving management, and protecting public safety. The Support Vector Machine (SVM) model, renowned for handling small samples, nonlinearity, and high-dimensional data, is well-suited for assessing WDN risks with limited failure data. However, it faces challenges such as difficulties with large datasets, selecting optimal kernel functions, and offering clear interpretability. To address these challenges and accurately assess pipeline risks, this study introduces an integrated CF-SVM model, combining the Certainty Factor (CF) model with SVM. The CF model, grounded in statistical theory, effectively manages uncertainties arising from multiple factors in pipeline failures. Results show the CF-SVM model outperforms standalone SVM and CF models, with an AUC of 0.92—improving accuracy by 17.95 % and 12.20 %, respectively. The model effectively allocates 71.31 % of faulty pipes to a smaller high-risk zone (22.96 %), enhancing both accuracy and regional applicability. Its application in real WDNs in China demonstrates its effectiveness in risk assessment and network safety management.