A study on pipe failure analysis in water distribution systems using logistic regression

Abstract This study presents a pipe failure analysis that aims to identify key factors affecting pipe failure in a pilot study area (PSA). For this purpose, pipe failures that occurred in the PSA were recorded for 2 years. Recorded pipe failures were integrated into the existing geographical information systems of the PSA and then visualized. Then, the logistic regression model that calculates failure probability with unbalanced and small- and medium-sized data was performed to analyze pipe failures. Pipe material, age, diameter, burst pressure, and average water pressure are the factors used in the study. The developed model achieved a success rate of 70.1%, suggesting that it can predict a considerable portion of pipe failures with a relatively satisfactory performance. Findings show that the failure probability of pipes with an age greater than 20 years is 1.675 times higher than that of pipes with an age less than 20 years. Moreover, polyvinyl chloride pipes are 5.265 times and high-density polyethylene pipes are 9.027 times more likely to fail than ductile pipes. Further research should delve into the impacts of the number of service connections, pipe lengths, and traffic load on pipe failures.