Statistical Modeling in Absence of System Specific Data: Exploratory Empirical Analysis for Prediction of Water Main Breaks

Abstract

The replacement of deteriorating distribution pipes is an important process for water utilities. It helps reduce capital spending on water main breaks and improves customer satisfaction. To assist with the development of an effective renewal plan, statistical models that forecast future breakage rates have been used to guide planning for asset management. However, this process is difficult for older utilities that lack readily available pipe network data. We examined whether accurate and useful predictive models can be built in the absence of pipe-feature data. Using the historical break record from a mid-Atlantic utility, two data sets at different spatial scales were created using publicly available demographic and environmental information. Empirical results suggest that although accuracy suffers from the lack of pipe-level details, it is still possible to create a model that provides useful information for prioritization of high-risk regions for management.