Projection of Water Demand and Sensitivity analysis of Predictors affecting Household usage in Urban Areas using Artificial Neural Network

Abstract

Maintaining a stable residential water supply is becoming a challenge as climate change persists to affect environmental conditions. Thus, the use of forecasting tools is suggested as they can detect complications in water management and operation. Artificial neural network (ANN) model was established in this study to analyze the influence of each selected predictors of water supply in Metro Manila which were identified considering the country's condition. The model used socioeconomic surveys and historical climatic data to train and validate the model. Internal model parameters such as the Levenberg-Marquardt algorithm and hyperbolic tangent sigmoid function was utilized. The developed model has a topology of 9-19-1 (input-hidden-output) and it yielded extremely high R-values 0.97013 and very low mean square error of 2.3463. The data were also used in sensitivity analysis to identify the degree of significance for each predictor that are known to affect water usage in urban areas. Among the selected predictors, the household income holds the highest impact on water demand with SI of 1.346599. The number of female adults has the highest significance among the other demographic variables with 1.215813. Rainfall and temperature data must be paid attention to as well as they are in the 3rd and 4th rank, respectively. The model could be used as a basis to plan for the future and to understand how much water the region will need to continuously sustain daily individual and economic activities.