Machine learning approaches for leak detection in water distribution systems: a comparative study

Mileiko, S; Karim, H; Balsamo, D

doi:10.1109/SAS65169.2025.11105198

Machine learning approaches for leak detection in water distribution systems: a comparative study

Lookup NU author(s): Dr Sergey Mileiko, Dr Domenico Balsamo

Downloads

Accepted version [.pdf]

Licence

This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).

Abstract

Leak detection in water distribution systems (WDS) is crucial for water conservation, as unnoticed leaks cause significant waste and infrastructure damage, leading to costly repairs. Traditional methods like visual inspections are accurate but labor-intensive and impractical for hard-to-reach pipes. Advancements in the Internet of Things (IoT) enable continuous WDS monitoring, collecting data on water flow, pressure, acoustic noise, and temperature without human intervention. At the same time, machine learning (ML) techniques provide the ability to analyze this data to detect and locate leaks. This study compares multiple ML techniques—support vector machines (SVM), decision trees, random forests, logistic regression, neural networks (NN), and k-means clustering—for leak detection in WDS. Their performance is assessed using simulated sensor data (water flow, pressure, acoustic noise, and temperature) from two EPANET-based WDS models representing small and large neighborhoods. Results indicate that random forest performs best in the small water network using only water flow data. At the same time, when additional input data (pressure, acoustic noise, and temperature) are considered in the large network, all the evaluated ML models demonstrate high performance.