<![CDATA[Clean water demand forecasting is essential for supporting production planning and distribution management in municipal water utilities. This study developed an integrated machine learning and rule-based Decision Support System (DSS) for forecasting clean water demand at the Lubuklinggau Municipal Water Utility. The study used 60 monthly observations from 2021–2025, which resulted in 58 observations after preprocessing and feature engineering. Random Forest Regression (RFR) was evaluated using five-fold TimeSeriesSplit and compared with Support Vector Regression (SVR) using the same temporal validation framework. The results showed that RFR outperformed SVR, achieving an overall out-of-fold R² of 0.7153, with lower prediction errors than the comparative model. Feature importance analysis indicated that water production, lagged demand, population, and Non-Revenue Water were among the most influential predictors. The final RFR model was subsequently retrained using all 58 observations to generate a baseline forecast for January–December 2026, with monthly clean water demand ranging from 675,563.98 to 691,313.64 m³. The forecast outputs were integrated into a rule-based DSS to classify demand conditions and generate corresponding operational recommendations. The proposed framework demonstrates the integration of temporally validated machine learning forecasting with interpretable rule-based decision support. However, the DSS remains a prototype and requires expert validation, user acceptance testing, and operational evaluation before implementation.]]>
Copyrights © 2026
