The quiet hum of a city at dawn masks a hidden crisis: water slipping away unseen. Every year, utilities worldwide lose billions of liters to leaks—water that never reaches homes, businesses, or the environment where it’s needed. But a breakthrough from Poland is changing the game.
Researchers led by Izabela Rojek from Kazimierz Wielki University in Bydgoszcz have developed a method to detect hidden leaks in water networks using a technique called Classification and Regression Trees (C&RT). Unlike complex black-box AI models, C&RTs work like a flowchart—simple, interpretable, and fast. By analyzing real-time data from sensors measuring pressure, flow, and consumption, the model can flag anomalies that signal a leak before it becomes a major problem.
“Our goal wasn’t just to detect leaks,” says Rojek, “but to give water utilities a tool they can trust and use every day—one that doesn’t require a data science PhD to understand.”
The results speak for themselves: the best-performing model achieved 86.15% accuracy in classifying leak conditions. That might not sound like perfection, but in a sector where every percentage point translates to millions of liters saved, it’s a game changer. For energy-intensive water systems, where pumping water uphill or across long distances consumes vast amounts of electricity, even small leaks add up to large energy waste. Fixing them isn’t just about saving water—it’s about cutting carbon emissions and operational costs at the same time.
What makes this approach particularly powerful is its integration into cyber-physical systems (CPS)—the backbone of smart water networks. By embedding C&RT models into real-time monitoring platforms, utilities can move from reactive repairs to predictive maintenance. Instead of waiting for a leak to surface as a burst pipe or a customer complaint, operators receive alerts based on data patterns, allowing them to act before damage escalates.
This isn’t just theory. Early pilots in European water utilities show that even modest improvements in leak detection can reduce non-revenue water by up to 15%, with corresponding drops in energy use. In an era where water and energy are increasingly intertwined—think desalination plants, wastewater recycling, and high-lift pumping—the ability to optimize both is a strategic advantage.
Rojek and her team are already looking ahead. “We’re not stopping at C&RT,” she notes. “Other decision tree variants could push accuracy even higher, and we’re exploring how to integrate these models with IoT sensors and cloud platforms for end-to-end smart water management.”
Published in *Applied Sciences* (Zastosowania Nauki in Polish), this work is a reminder that innovation doesn’t always require reinventing the wheel—sometimes, it’s about sharpening the tools we already have. For an industry under pressure to do more with less, that clarity is priceless.