Classification of Anomalous Water Meters Using Machine Learning and Spatial Data

Abstract

Water loss management is a crucial task in water supply services. Water loss occurs due to physical leakage in the pipeline system and commercial losses from malfunctioning water meters, leading to revenue loss and reduced consumer confidence for the Provincial Waterworks Authority (PWA). This study aims to develop a machine learning (ML) model to classify water meters prone to malfunction using XGBoost, LSTM, and a Stacking Ensemble approach. The research utilizes data on meter age, customer type, brand, 49-month historical consumption records, and spatial data such as the number of pipe repair points and the distance to repair sites in the PWA Rangsit branch, Pathum Thani Province. The results show that the Stacking Ensemble model incorporating spatial data achieves the highest accuracy 90.03%. This study enables PWA to accurately detect malfunctioning water meters, reduce water loss, enhance fairness in water billing, and expand model applications to other areas in the future.