Interpretable AI for Short-Term Water Demand Forecasting

Machine learning models such as artificial neural networks (ANNs) are becoming increasingly popular in short-term water demand forecasting. This is because, despite their lack of interpretability, ANNs are able to capture complex interactions between explanatory variables and water consumption better than a traditional time series analysis or simple linear regression. In this work, we forecast the hourly water demand of ten operational district metered areas using optimal trees, a machine learning model which has been shown to combine the interpretability of regression approaches and the accuracy of ANNs. We show that, compared to existing water demand forecasting models, optimal trees offer valuable insights without sacrificing predictive or computational performance.