Machine Learning-Based Estimation of Sewage Treatment Facility Capacity and Design Adequacy: A Case Study in Korea

Accurate estimation of regional sewage generation is essential for designing reliable and resource-efficient treatment facilities. This study developed an ensemble machine-learning framework to estimate annual sewage generation (SG) as the primary output variable, using a combination of demographic, socioeconomic, and environmental indicators across multiple regions in Korea. The proposed Voting Regressor model, trained using data from four highly urbanized regions (Regions A–D), effectively captured nonlinear interactions among variables such as population, business establishments, economically active population, rainfall, and gross regional domestic product (GRDP). A generalization test on an unseen region (Region E) confirmed the model’s robustness and transferability, demonstrating that the framework can reliably adapt to regions with different demographic and industrial characteristics. Comparative analyses showed that the model outperformed both the Random Forest and the conventional per capita unit-load (GU) method in terms of the coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE). SHAP (Shapley Additive Explanations) analysis further identified business establishments and GRDP as the dominant contributors to sewage generation. Moreover, model-based capacity estimations incorporating a 20% safety factor are closely aligned with actual facility capacities, revealing that conventional design standards often apply excessively conservative margins. The findings demonstrate that the proposed machine learning framework can quantitatively assess design adequacy and prevent structural overestimation while maintaining sufficient operational reserves. This data-driven approach provides an interpretable and adaptable foundation for future sewage infrastructure planning and rational capacity design under evolving socioeconomic and environmental conditions.