Abstract

The increasing deployment of smart water meters has transformed water distribution systems by enabling high-resolution, near real-time monitoring of customer level water consumption. In water scarce regions such as the Middle East and North Africa (MENA), and particularly in the United Arab Emirates (UAE) where potable water production relies predominantly on energy intensive desalination, these data streams present a critical opportunity to reduce Unaccounted-for Water (UFW), also referred to as Non-Revenue Water (NRW). UFW arising from physical leaks, meter malfunctions, unauthorized connections, and abnormal consumption patterns imposes significant economic, operational, and environmental burdens on water utilities (World Bank, 2016; World Bank, 2020). However, the complexity, noise, and temporal variability inherent in smart water meter data limit the effectiveness of conventional rule-based and threshold-based detection approaches. This research proposes a machine learning–based anomaly detection framework for identifying UFW related events using smart water meter time-series data. The study evaluates supervised, unsupervised, and semi-supervised learning techniques suitable for real-world utility environments where labeled leak and tampering events are often scarce (Mounce et al., 2018; Romano et al., 2017). Temporal consumption modeling and feature extraction are combined with anomaly detection algorithms to identify deviations from learned normal behavior that may indicate leaks, continuous flow events, or metering irregularities. The analytical pipeline includes data preprocessing, feature engineering, model training, and comparative evaluation using real-world smart meter datasets. Model performance is assessed using standard machine learning metrics such as precision, recall, and F1 score, alongside utility oriented indicators including false alarm rates, detection timeliness, and estimated volumetric water savings. By explicitly linking analytical performance to operational outcomes, this research demonstrates how machine learning can support improved water loss management, reduced energy consumption, and enhanced system resilience in water stressed regions. The findings contribute both to the academic literature on smart infrastructure analytics and to practical decision making for utilities seeking to improve the sustainability of urban water distribution systems.

Publication Date

9-2026

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Department, Program, or Center

Office of Graduate Studies

Advisor

Ehsan Warriach

Comments

This thesis has been embargoed. The full-text will be available on or around 4/20/2027.

Campus

RIT Dubai

Download

Share

COinS