Water use prediction by radial and feed-forward neural nets

In this study, applicability of feed-forward and radial-basis neural networks for monthly water consumption prediction from several socio-economic and climatic factors affecting water use is investigated. A data set including a total of 108 data records is divided into two subsets: training and testing. Firstly, the models based on a single input variable are trained and tested by feed-forward and radial methods and feed-forward and radial performances of the models are compared. Then, the models based on multiple input variables are constructed according to performances of the models based on a single input variable. The performances of feed-forward and radial models in training and testing phases are compared with the observations and the best-fit model is identified. For this purpose, several criteria such as normalised root mean square error, efficiency and correlation coefficient are calculated for all models. Subsequently, the best-fit models are also trained and tested by multiple linear regression for comparison. The results indicated that feed-forward and radial methods can be applied successfully for monthly water consumption prediction.