Comparative Analysis of Ensemble Learning Models for Predicting the Sludge Production in Wastewater Treatment Plants

Accurate prediction of sludge production is critical for the efficient operation, cost optimization, and strategic planning of wastewater treatment plants (WWTPs). This study investigated the application of three advanced ensemble machine learning models, including Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), to predict daily sludge production at the Southwest Tehran WWTP. The modeling framework incorporated multicollinearity reduction using Spearman correlation analysis and hyperparameter optimization via Optuna. Model performance was evaluated using the coefficient of determination (R²), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE) across both training and test datasets. Among the developed models, XGBoost outperformed the others, demonstrating the highest predictive accuracy on unseen data. Feature importance analysis using the XGBoost model identified influent flow rate and return sludge flow as the most influential predictors. These findings confirm the effectiveness of gradient boosting techniques for modeling complex operational data in WWTPs and underscore their potential for real-time application and integration into intelligent decision-support systems.

Keywords:

Ensemble Learning , Feature Importance , Sludge Production , Tree-based Machine Learning Algorithms , Wastewater Treatment