Conventionally, multiple linear regression methods have been used to predict water losses (a proportion of which is real losses) some weeks or months ahead, based upon various weather parameters. This paper describes the development of an alternative method to predict water losses using random forests and compares model performance with linear regression using a case study approach from one water utility. It suggests that a random forest approach can significantly improve the ability to predict water losses based on readily available covariates. Further validation work on holdout data is recommended to ensure the model is not over-fitted to the learning set.

You do not currently have access to this content.