Fig. 2
From: Short-term water demand forecasting using hybrid supervised and unsupervised machine learning model
Forecasting methodology flowchart. A flowchart of the steps developing, calibrating, and validating the models is shown. The historical water demand raw data is pre-processed, where missing and erroneous data are cleaned and imputed. Then, processed data is fed into the SARIMA model, whereas it is divided into two groups, target data and input data for the hybrid model. For the hybrid model, the target data is fed into the SOM model and the input data is sent directly to the RT model. In the SOM model, the target data is clustered and the output cluster number accompanied with the target data is added to the input data in the RT model. At this point, the RT model performs the prediction t time steps ahead. After predicting the target, the performance of the model is assessed (i.e. compared to the held back water demand data). If the performance is satisfactory, the model is implemented to forecast t time steps ahead. However, usually the desired performance cannot be attained from the first trial. In this case, more neurons can be added to the SOM model and/or more leaves and folds can be added to the RT model