Predicting the Economic Viability of Rainwater Harvesting Systems Using Machine Learning: Insights from 223 Buildings in Guangzhou, China

Urban rainwater harvesting (RWH) systems play a vital role in sustainable water management, yet their widespread adoption is often impeded by economic feasibility concerns. Traditional approaches struggle to capture the nonlinear, multivariable nature of RWH economic analysis. This study addresses these limitations by developing a machine learning-based framework to predict the economic viability of decentralized rooftop RWH systems in Guangzhou, China. Using data from 223 buildings, four advanced machine learning algorithms—Random Forest (RF), AdaBoost, Light Gradient Boosting Machine (LGBM), and Multilayer Perceptron (MLP)—were evaluated. Among them, the RF model demonstrated the highest performance (R² > 0.97, RMSE < 0.04, MAE < 0.03). The model was further refined using Bayesian optimization to improve predictive accuracy. Key findings highlight the catchment-to-floor area (C/A) ratio as the most influential determinant of economic viability, followed by building height and water demand. This research not only outperforms traditional methods by reducing prediction errors by more than threefold but also provides valuable insights for urban planners and policymakers in sustainable water management. Furthermore, it underscores the potential of machine learning in tackling complex urban challenges and sets the stage for future studies incorporating broader geographic and climatic datasets.