Probabilistic Water Demand Forecasting Using Quantile Regression Algorithms

Machine and statistical learning algorithms can be reliably automated applied at scale. Therefore, they constitute a considerable asset for designing practical forecasting systems, such as those related to urban water demand. Quantile regression are machine that provide probabilistic forecasts in straightforward way, have not been so far demand forecasting. In this work, we fill gap, thereby proposing new family of algorithms. We further extensively compare seven from one-day ahead settings. More precisely, five individual quantile (i.e., the regression, linear boosting, generalized random forest, gradient boosting neural network algorithms), their mean combiner median combiner. The comparison is conducted by exploiting large flow dataset, well several types hydrometeorological time series (which considered exogenous predictor variables setting). results mostly favour algorithm, probably due presence shifts (and perhaps trends) series. combiners also found skillful.