Automating Water Capital Activities Using Naïve Bayes Classifier with Supervised Learning Algorithm

Abstract

Municipal governments have the responsibility to provide safe drinking water to residents. Maintaining water infrastructure systems to keep a certain level of service is a vital service. It is possible by assessing all assets and planning capital work activities to renew and renovate the existing assets. The municipalities prioritize the capital activities of their infrastructure and are required to optimize their available resources. Past studies confirmed due to several complexities and imperfections of the available water network data, there is a need for a comprehensive multicriteria database to prioritize pipe capital plan decisions based on engineering expert judgment. This database must include information about water pipe physical condition and performance up to an acceptable level of service and criticality based on the water pipe location. In addition, the lack of standard regulatory requirements due to incomplete condition, criticality and performance assessment of the entire Municipal Water Network (MWN) leads to bias and undefendable engineering judgment. Although several pipe prioritization models have been developed and published in the literature, no comprehensive multi-decision criterion model is available to date, including the pipe segment condition, performance, and criticality. In this research, a novel Priority Action Number (PAN) is developed and parameterized based on pipe segment condition, performance and criticality. An automated Naïve Bayes Classifier (NBC) with a supervised machine learning model is proposed for consistent, defensible and personnel independence ranking of existing water pipe condition, performance, and criticality of all water pipes through MWN. This methodology automates the capital activities decision-making process. The research presents and develops a prioritizing approach for the MWN capital activities and aids in selecting assistive technology for rehabilitation and renewal capital activities. The developed model is applied to the City of London MWN database in a Geographical Information System (ArcGIS) database to validate and verify the model. The multi-level classifier model classified and assigned a capital work activity to all pipes in the City of London MWN. The presented multi-level NBC with a supervised learning algorithm replicates the expert's opinion and engineering judgement. Through NBC supervised machine learning algorithm, the capital project decision-making process is automated. This methodology will add consistency and defensibility to capital programs. Using this algorithm can help utility save money by automating industry best practices and optimizing long-term decisions about the order in which pipes need to be staged into capital works programs.