Anomaly Detection in Multifamily Water Usage
Overview
As a business intelligence analyst in the multifamily housing industry, I developed a billing system automation and an early anomaly detection system.
Details
In 2008, during the economic downturn, the company assumed responsibility for utility billing. With nearly 500 units spread across four complexes, the initial billing process was manual and prone to errors. Often, when bills were high, property managers gave credit for the bills they should have been collecting. The problem escalated over time, resulting in less than 70% of the water bills being collected.
Problem Statement
Despite a large investment in sophisticated submeters for four different complexes, the company is still struggling to collect on water bills. A large gap remains between what each complex pays to the water utility company and the amount it collects earmarked for that purpose.
Investigation and Root Cause Analysis
By reviewing historical meter data and observing usage spikes, I discovered a pattern: isolated, sharp consumption increases. In one case, a continuously running toilet in a vacant unit turned out to be the culprit. The hypothesis: toilet flapper malfunction.
Solution: Early Warning System
Instead of waiting for monthly billing cycles, I proposed a proactive approach—weekly anomaly detection. The goal was to catch abnormal water usage before residents were billed.
Data Preparation
Collected over 12 months of historical meter readings
Enriched the data with units metadata
Stratified data by apartment
Removed statistically insignificant factors using the Kruskal-Wallis test