Concurrent trends of urbanization and population growth in Brazil can exert high pressure on the (already degraded) environment. In the city of São Paulo, in particular, there is a clear trend towards verticalization of real estate, increasing population density. To attend demands due to this rapid change in a particular area, water consumption, it is necessary to understand the aspects related with domestic water demand. The main objective of this study is to analyze the monthly water consumption in high-rise residential properties, and investigate the descriptive power of building related variables using machine learning. For such, real consumption data from the past three years (provided by the water and sewage company Sabesp) were obtained, along with two databases containing detailed information on high-rise apartment buildings in the city of São Paulo. After a meticulous integration of these databases, reliable information were obtained for 3,299 high-rise buildings, totalizing 276,670 apartments, described by 21 variables. One potential weakness in commonly used estimates (e.g., demographic, financial) is that they may be outdated or biased. In contrast, the physical characteristics of buildings are easily verifiable, and simple to obtain. The study’s hypothesis is that relying solely on the building features may preserve a similar descriptive power, while eliminating uncertainties and biases. A contribution of this study is the estimation of the monthly consumption per unit, which can be used for modeling urban water distribution systems. In the experiments carried out, fourteen different regression algorithms for consumption prediction are investigated, and the predictive performance of the induced models is comparable with similar studies that use building characteristics alongside population estimates and water/sewage features in the building, partially confirming the research hypothesis.
