Scholarlink

<p>This study proposes an interpretable machine learning framework for predicting effluent quality in the Lake Manzala Hybrid Constructed Wetlands (HCWs), Egypt, with specific emphasis on biochemical oxygen demand (BOD) and chemical oxygen demand (COD). Accurate prediction of BOD and COD is essential for evaluating organic matter removal, supporting wastewater reuse, and improving operational decision-making in constructed wetland systems under water-scarcity conditions. The study uses real monitoring data collected from the Lake Manzala HCWs under two hydraulic loading regimes: 50&#xa0;m<sup>3</sup>/day and 83.3&#xa0;m<sup>3</sup>/day. Water quality measurements were obtained from the inlet, intermediate sampling points, and outlet of the wetland flow path, while outlet BOD and COD concentrations were used as the target variables. A hybrid linear regression–gradient boosting (LR–GB) model was developed by first capturing the dominant linear relationships using linear regression (LR) and then modeling the remaining nonlinear residual patterns using gradient boosting (GB). The proposed LR–GB model was compared with several conventional regression models, including random forest (RF), K-nearest neighbors (KNN), decision tree (DT), Bayesian ridge regression (BR), and support vector regressor (SVR), using mean squared error (MSE), mean absolute error (MAE), median absolute error (MedAE), root mean squared error (RMSE), and mean absolute percentage error (MAPE) as evaluation metrics. The results demonstrate that the proposed LR–GB framework achieved the lowest prediction errors across the investigated BOD and COD cases. For COD prediction, the proposed model achieved MAPE values of 3.87% and 3.46% under the 50&#xa0;m<sup>3</sup>/day and 83.3&#xa0;m<sup>3</sup>/day discharge conditions, respectively, outperforming all benchmark models. These findings confirm that the LR–GB framework can effectively combine interpretability and nonlinear predictive capability for effluent quality modeling. The developed model provides a practical decision-support tool for wastewater treatment operators and water-resource managers, contributing to improved effluent quality control, safe wastewater reuse, and sustainable water management in arid and semi-arid regions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of effluent biochemical and chemical oxygen demand in Lake Manzala Hybrid Constructed Wetlands, Egypt using a linear regression-gradient boosting framework

  • 期刊论文

  • 发表日期：

    2026年7月24日

查看全文
  • Ahmed Fahim,
  • Ahmed M. Elshewey

摘要

This study proposes an interpretable machine learning framework for predicting effluent quality in the Lake Manzala Hybrid Constructed Wetlands (HCWs), Egypt, with specific emphasis on biochemical oxygen demand (BOD) and chemical oxygen demand (COD). Accurate prediction of BOD and COD is essential for evaluating organic matter removal, supporting wastewater reuse, and improving operational decision-making in constructed wetland systems under water-scarcity conditions. The study uses real monitoring data collected from the Lake Manzala HCWs under two hydraulic loading regimes: 50 m3/day and 83.3 m3/day. Water quality measurements were obtained from the inlet, intermediate sampling points, and outlet of the wetland flow path, while outlet BOD and COD concentrations were used as the target variables. A hybrid linear regression–gradient boosting (LR–GB) model was developed by first capturing the dominant linear relationships using linear regression (LR) and then modeling the remaining nonlinear residual patterns using gradient boosting (GB). The proposed LR–GB model was compared with several conventional regression models, including random forest (RF), K-nearest neighbors (KNN), decision tree (DT), Bayesian ridge regression (BR), and support vector regressor (SVR), using mean squared error (MSE), mean absolute error (MAE), median absolute error (MedAE), root mean squared error (RMSE), and mean absolute percentage error (MAPE) as evaluation metrics. The results demonstrate that the proposed LR–GB framework achieved the lowest prediction errors across the investigated BOD and COD cases. For COD prediction, the proposed model achieved MAPE values of 3.87% and 3.46% under the 50 m3/day and 83.3 m3/day discharge conditions, respectively, outperforming all benchmark models. These findings confirm that the LR–GB framework can effectively combine interpretability and nonlinear predictive capability for effluent quality modeling. The developed model provides a practical decision-support tool for wastewater treatment operators and water-resource managers, contributing to improved effluent quality control, safe wastewater reuse, and sustainable water management in arid and semi-arid regions.