Data-driven models for sewer blockage prediction

Water and waste utilities companies are under pressure to deliver a better service with a lower cost for consumers. It is important for these companies to understand all the factors that influence sewer blockages and be able to control them by prioritizing proactive strategies. These companies are keen to find solutions to reduce the occurrences of blockages on their wastewater network, which furthermore will help reduce maintenance costs, customer and environmental impact. This paper presents a data mining (DM) base approaches to predict Sewer blockages using absolute levels in EDMs (event duration monitors) and SLMs (sewer level monitors). Three different DM approaches are used (Decision Trees, Logistic Regression, and Random forest) to build the prediction models. The accuracy of these models is evaluated using real datasets containing blockage incident records for one of the biggest water and waste services providers in the UK, which will be denoted by Provider x in this research.