Abstract
Developments in data mining techniques have significantly influenced the progress of Intelligent Water Systems (IWSs). Learning about the hydraulic conditions enables the development of increasingly reliable predictive models of water consumption. The non-stationary, non-linear, and inherent stochasticity of water consumption data at the level of a single water meter means that the characteristics of its determinism remain impossible to observe and their burden of randomness creates interpretive difficulties. A deterministic model of water consumption was developed based on data from high temporal resolution water meters. Seven machine learning algorithms were used and compared to build predictive models. In addition, an attempt was made to estimate how many water meters data are needed for the model to bear the hallmarks of determinism. The most accurate model was obtained using Support Vector Regression (8.9%) and the determinism of the model was achieved using time series from eleven water meters of multi-family buildings.