Domestic Water Consumption Forecasting With Sociodemographic Features Using ARIMA and ARIMAX: A Case Study in Malaysia

Abstract

Domestic water consumption in Malaysia exceeds the World Health Organisation’s (WHO) recommendation and the target by National Water Services Commission (SPAN). This could cause problems to the future water supply. Domestic water consumption forecasting for each state proposed in this paper could be used to formulate targeted strategies for water management and conservation awareness. Time series annual data between 2010 to 2019 obtained from Malaysian government portals were used to forecast domestic water consumption based on its past values and socio-demographic data. Variance Infation Factor was used to remove collinear variables, while non-collinear features were selected based on high Pearson and Spearman correlations and XGBoost and AdaBoost permutation-based feature importance. It was found that percentage of the population above 64 years old, gender ratio, and mean household size were the important features for all states. These features were included in ftting ARIMAX models in forecasting domestic water consumption one year ahead for each state. However, time series forecasting modelling results show that ARIMA models using only past data for domestic water consumption for each state have better accuracy than ARIMAX models, except for Kedah and Perak. The need to consider important features diferently for each state, linearity assumption of exogenous variables and insufcient data could be the reasons for lower R2 values and higher RRMSE and MAPE for ARIMAX models. Thus, the forecasts for the year 2020 were obtained using ARIMAX models for the two states and ARIMA models for others. The forecasted values show that only three states would meet the target by SPAN, while others were expected to exceed it. This shows that more eforts are needed to increase water conservation awareness, focusing on the states with high forecasted domestic water consumption.