Implementing Machine Learning Algorithms for Pipe Failure Analysis in Water Distribution Systems

Abstract Pipe failures in water distribution systems occur due to various physical and environmental factors, and their analysis is essential for improving water loss control and asset management strategies. This study provides an innovative application of machine learning techniques to analyze pipe failures in a pilot study area (PSA) in Antalya, Türkiye, where failure data were recorded over a seven-year period. The factors considered in this study include pipe material, age, diameter, and length. Eight machine learning techniques, including Naïve Bayes, artificial neural networks, logistic regression, XGBoost, LightGBM, CatBoost, random forest, and support vector classification, were applied for pipe failure analysis. According to the findings, XGBoost achieved the highest prediction accuracy with an Area Under Curve (AUC) value of 0.99. The results further identified pipe length as the most influential factor in pipe failures, while PVC pipes were found to be the most prone to failure in the PSA. This comprehensive analysis not only provides valuable insights into pipe failure analysis but also offers an effective tool for optimal management of water distribution systems. Future studies should incorporate additional factors, such as the number of service connections and traffic load, to enhance failure probability assessments and inform optimal pipe replacement strategies.