Low-Resolution Monthly Billing Data for Residential Water Demand Management: A Profile-Based Approach to Apartment-Level Characterization and Forecasting

Monthly water-utility billing records are widely available, yet their potential to support apartment-level characterization and forecasting of residential water consumption in multifamily buildings remains underexplored. This study evaluates an integrated and interpretable framework using monthly billing data from 320 apartments across four middle-income multifamily buildings in Brazil from 2018 to 2024. The procedure combined preprocessing, outlier and missing-data treatment, normalization, temporal analysis, Dynamic Time Warping (DTW) + TimeSeriesKMeans clustering, sensitivity analysis, and forecasting using naïve, seasonal naïve, Exponential Smoothing State Space (ETS), Structural State Space Model (SSM), and Seasonal Autoregressive Integrated Moving Average (SARIMA) models. Profile-based stratification was validated against a non-clustered reference. Clustering identified two profiles: a minority profile with lower typical consumption, greater variability, and higher intermittency, and a majority profile with higher typical consumption and greater temporal regularity. The missing-data sensitivity analysis showed 94.06% agreement between scenarios, although the minority profile was more sensitive to the adopted strategy. SSM performed best in six of eight series, with symmetric Mean Absolute Percentage Error (sMAPE) values from 4.87% to 28.16%. In contrast, the naïve benchmark was superior in two series with strong short-term persistence. Profile-based stratification improved performance in five series but worsened it in three, mainly when clustering reduced heterogeneity or aggregation bias. These results show that existing billing records can support reproducible apartment-level characterization and forecasting where high-resolution monitoring is limited, and that predictive performance depends more on profile regularity than on model complexity.