ABSTRACT
Ensuring efficient water distribution requires advanced predictive strategies to enable the urban centres across the world address huge problems in water resource management. This study employs machine learning algorithms to build predictive models that can forecast accurately water tanker demands in Hyderabad by using datasets obtained from Telangana government's open data platform. We used four popular ML models for this, which included Lasso Regression, K-Nearest Neighbors (KNN), XGBoost and Random Forest. For each algorithm, the evaluations were done based on Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) where KNN performed best showing promising results which had RMSE 0.62 and MAE 0.03 compared to other models. Consequently, KNN with its lower error metrics is a more dependable and accurate means for municipal authorities to predict water needs and hence offers an avenue for them to establish informed and efficient resource allocation decisions. The study also compares how models differ significantly in their performances as well as what practical implications need be considered when selecting suitable predictive models for improving urban water management systems.
