Increasing global populations, changing water consumption patterns and climate change are continuously putting a strain on available water resources. Identifying these changes in consumption patterns and future demand requirements through water demand forecasting plays a key role in the allocation, planning, and management of urban water resources and infrastructure. In the last decades, advances in statistics, machine learning, and artificial intelligence techniques have shown promising results in predicting future water demands. This study aims to forecast hourly water demand, while factoring the effects of other exogenous features on consumption patterns using machine learning techniques for Telford Borough, a predominantly residential town in Pennsylvania, US. In our study, we have compared the predictive abilities of the support vector machine and the random forest regression models in forecasting water demand using past demand, and temporal and climatic features. The results demonstrated the practicability of applying machine learning solutions to water management with Random Forest being the best model, and temporal features being the most important in our exploratory variables. We also conclude that these predictive models can be used for near real-time forecasting to guide allocation and planning decisions of water resources.
