A systematic review of machine learning and signal processing techniques for water pipe leakage prediction

The efficient management of water distribution systems is a critical global challenge, primarily due to the escalating volume of Non-Revenue Water (NRW) caused by undetected pipe leakages. This systematic review explores the integration of machine learning (ML) and signal processing (SP) techniques as a robust solution for leakage prediction, viewed through the interdisciplinary lenses of electronics engineering, instrumentation, and computer programming. Following a thematic and regional framework, the review analyzes technical, economic, instrumentation-based, and operational barriers hindering the adoption of these smart technologies. A systematic search of literature from 2017 to 2026 was conducted using databases such as IEEE Xplore, ScienceDirect, and Google Scholar, focusing on the synergy between hardware prototypes and software algorithms. The findings reveal a complex interplay of barriers: technical challenges involve signal attenuation in non-metallic pipes and the need for noise-resilient filtering; economic constraints stem from high capital costs of high-fidelity sensors and regional socio-economic disparities; and programming barriers include the computational demand of deep learning models on embedded edge devices. Regional analysis highlights significant disparities, with high-income regions advancing toward digital twins while developing regions in Africa and Asia utilize low-cost, open-source rapid prototyping. The review concludes that the future of leakage prediction lies in the standardization of low-cost instrumentation and the deployment of "TinyML" for real-time edge analytics to ensure global water sustainability.