Investigating the accuracy of multivariate regression and ARIMA models in predicting water demand (Case Study: Mashhad city)

Awareness of water demand is of particular importance for its policy in urban management. Predicting water demand in the future will allow managers to take the necessary measures regarding sustainable water supply, given the constraints and crises ahead. The purpose of this study is to compare multivariate regression and ARIMA models to predict water demand in Mashhad. In this study, first, the main variables affecting water demand including rainfall, temperature, and population were determined and then the required data were collected from relevant organizations. After checking the accuracy and homogeneity of the data, water demand was estimated by two models of multivariate regression and ARIMA. The results showed that for data training, the determination coefficient and Nash coefficient were 0.81 and 0.77 respectively for the multivariate regression model and 0.77 and 0.73 for the ARIMA model, and the same coefficients for training data are respectively 0.86 and 0.87 for the multivariate regression model and 0.79 and 0.73 for the ARIMA model. Therefore, the multivariate regression model has a better performance in terms of prediction due to the high determination coefficient and being in the very good category of the Nash coefficient, and it can be used as an acceptable method for predicting water demand compared to the ARIMA model.