Water demand forecasting: a comparative analysis with machine learning techniques

Código

XXVI-SBRH0224

Autores

Verônica Batista de Oliveira, Elizabeth Pauline Carreño Alvarado, Gilberto Reynoso Meza

Tema

STE140 - Análise de Sistemas de Abastecimento de Água

Resumo

The accelerated population growth, unplanned urbanization, and climate change have increased the need for efficient and evidence-based water management. In this context, machine learning (ML) techniques have emerged as a promising tool for forecasting phenomena and supporting decision-making processes. This study aims to compare and analyze methods for predicting urban water demand over short- and long-term horizons using machine learning approaches. The data used originate from cities in northeastern Italy and were made available through the international challenge "Battle of Water Demand Forecasting" held as part of the WDSA-CCWI conference. The development follows the CRISP-DM methodology, including exploratory analysis, modeling, and quantitative validation. Strong seasonality and autocorrelation were identified in the data, particularly in daily and weekly patterns. Based on these findings, various algorithms were implemented and compared in terms of predictive performance, using standardized metrics. The results demonstrate the feasibility of applying ML-based models to support intelligent water management in urban contexts, contributing to sustainability in the planning and operation of water systems.