Estimation of Community Water Consumption Using Multivariate Ensemble Approach.

Water consumption prediction is an important application of Time Series forecasting, as it has various real-life applications. Judicial usage and optimal management of water are imperative and accurate forecasting is a key component in achieving those goals. This paper explores a novel multivariate ensemble approach for forecasting future water consumption, using a combination of Machine Learning and Statistical Modelling. The models used in this project are SARIMAX, Prophet and XGBoost. The models are trained individually on the data and their respective results are then used to predict the final output. The final ensemble model can forecast up to 36 months of data into the future. The model was able to output results with scores of 360.075 MSE, 18.976 RMSE, 15.814 MAE, and 0.027 MAPE.