Detection of unauthorized consumption in water supply systems: A case study using logistic regression

Abstract

This study proposes characterizing unauthorized water consumption and fitting multiple logistic regression models for detecting illegal connections. The research used monthly water consumption data from 42,131 single-family homes in Joinville, southern Brazil. The obtained models achieved accuracies ranging from 83% to 100% for detecting three types of fraud (unauthorized reconnection of a disconnected water service line, bypass, and direct connection). The results showed previous minor violations, tampering with water meters and stands, and non-compliance of meter installation with utility standards are related to illegal connections.