Advanced Predictive Models for the Smart Management of Drinking Water Consumption

Abstract

Accurate prediction of drinking water consumption is a key challenge for utility providers, especially in contexts where manual metering remains the predominant method. In many regions of Argentina, cooperatives still rely on these records, which leads to billing errors, hinders operational planning, and creates conflicts with users. This work addresses this issue by evaluating and comparing three forecasting methodologies: a classic statistical method (Moving Averages), an ensemble machine learning model (Random Forest), and a specialized time-series procedure (Prophet). The results show that machine learning-based models, such as Random Forest, significantly reduce prediction errors compared to traditional methods. It is concluded that the adoption of these tools provides valuable insights to optimize strategic decision-making in cooperative water management.